Datasets:

shuvomonowar00
/

BTUMQA

qa_id string	split string	slice_id int64	unique_id string	image_id int64	question string	answer string	answer_type string	question_family string	template_id string	paraphrase_id string	dataset_branch string	question_style string	difficulty_level string	ambiguity_flag string	region_target_primary string	region_target_secondary string	signal_gap_bucket string	decision_rule_id string	label_provenance string	candidate_keep_reason string	modality_context string	tumor_slice_csv int64	edema_area_pixels int64	ncr_net_area_pixels int64	enhancing_area_pixels int64	tumor_area_pixels int64	context_area_pixels int64	global_area_pixels int64	edema_uncertainty float64	ncr_net_uncertainty float64	enhancing_uncertainty float64	tumor_uncertainty float64	context_uncertainty float64	global_uncertainty float64	edema_weight float64	ncr_net_weight float64	enhancing_weight float64	tumor_weight float64	context_weight float64	global_weight float64	phase2a_token_file string	slice_index_in_token_file int64	phase2c_file string	slice_index_in_ugtm_file int64	mc_dropout_passes int64
BTUMQA225K_24716f3aeba265	train	49	00000_049	49	What is the confidence-qualified presence status of edema in this slice?	not_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_01	confidence_qualified_presence_paraphrase_01	btumqa_225k	confidence_qualified	easy	no	edema	null	wide_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	4	0	0	4	378	13,050	0.045768	0	0	0.045768	0.014483	0.001143	0.954232	1	1	0.954232	0.985517	0.998857	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	0	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	0	20
BTUMQA225K_f7814293977158	train	49	00000_049	49	What is the confidence-qualified extent of the whole tumor in this slice?	none	classification	confidence_qualified_extent	confidence_qualified_extent_template_01	confidence_qualified_extent_paraphrase_01	btumqa_225k	confidence_qualified	easy	no	tumor	null	wide_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	4	0	0	4	378	13,050	0.045768	0	0	0.045768	0.014483	0.001143	0.954232	1	1	0.954232	0.985517	0.998857	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	0	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	0	20
BTUMQA225K_153fc861d55689	train	50	00000_050	50	How large is the uncertainty gap between enhancing tumor and the whole tumor region?	moderate_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_01	uncertainty_gap_bucket_paraphrase_01	btumqa_225k	bucketed	medium	no	enhancing	tumor	moderate_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_20faaba07d208b	train	50	00000_050	50	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	edema	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_9d20b7d9b0308a	train	50	00000_050	50	Between edema and enhancing tumor, which region is more reliable for reasoning?	edema	classification	more_reliable_region	more_reliable_region_template_01	more_reliable_region_paraphrase_01	btumqa_225k	comparative	medium	no	edema	enhancing	moderate_gap	more_reliable_region_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_ad3a6be227473a	train	50	00000_050	50	Between edema and enhancing tumor, where is the uncertainty higher?	enhancing	classification	more_uncertain_region	more_uncertain_region_template_03	more_uncertain_region_paraphrase_03	btumqa_225k	comparative	medium	no	edema	enhancing	moderate_gap	more_uncertain_region_v1	phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_c4c45b1d3165f7	train	50	00000_050	50	How large is the reliability gap between enhancing tumor and the whole tumor region?	moderate_gap	classification	reliability_gap_bucket	reliability_gap_bucket_template_01	reliability_gap_bucket_paraphrase_01	btumqa_225k	bucketed	medium	no	enhancing	tumor	moderate_gap	reliability_gap_bucket_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_cfba1c148f3b42	train	50	00000_050	50	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	42	45	92	179	892	13,409	0.053225	0.080099	0.060797	0.063873	0.016343	0.001996	0.946775	0.919901	0.939204	0.936127	0.983657	0.998004	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	1	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	1	20
BTUMQA225K_0191f290e4cad2	train	51	00000_051	51	Between enhancing tumor and the whole tumor region, where is the uncertainty higher?	tumor	classification	more_uncertain_region	more_uncertain_region_template_03	more_uncertain_region_paraphrase_03	btumqa_225k	comparative	easy	no	enhancing	tumor	wide_gap	more_uncertain_region_v1	phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	55	74	263	392	1,056	13,699	0.042999	0.073363	0.028725	0.039154	0.009119	0.001858	0.957001	0.926637	0.971275	0.960846	0.990881	0.998142	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	2	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	2	20
BTUMQA225K_598044a81b1c08	train	51	00000_051	51	Between enhancing tumor and the whole tumor region, which token is more reliable?	enhancing	classification	more_reliable_region	more_reliable_region_template_03	more_reliable_region_paraphrase_03	btumqa_225k	comparative	easy	no	enhancing	tumor	wide_gap	more_reliable_region_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	55	74	263	392	1,056	13,699	0.042999	0.073363	0.028725	0.039154	0.009119	0.001858	0.957001	0.926637	0.971275	0.960846	0.990881	0.998142	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	2	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	2	20
BTUMQA225K_817d460816e5f3	train	51	00000_051	51	Which tumor subregion is strongest after uncertainty-aware adjustment?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_02	dominant_region_under_uncertainty_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	edema	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	55	74	263	392	1,056	13,699	0.042999	0.073363	0.028725	0.039154	0.009119	0.001858	0.957001	0.926637	0.971275	0.960846	0.990881	0.998142	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	2	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	2	20
BTUMQA225K_d958aa40f2a509	train	51	00000_051	51	Is the uncertainty difference between edema and the whole tumor region close, moderate, or wide?	moderate_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_02	uncertainty_gap_bucket_paraphrase_02	btumqa_225k	bucketed	medium	no	edema	tumor	moderate_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	55	74	263	392	1,056	13,699	0.042999	0.073363	0.028725	0.039154	0.009119	0.001858	0.957001	0.926637	0.971275	0.960846	0.990881	0.998142	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	2	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	2	20
BTUMQA225K_de9133a92f3f0f	train	51	00000_051	51	Does this slice show an ambiguity-sensitive competition between edema and necrotic or non-enhancing tumor core?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_03	ambiguous_subregion_pair_paraphrase_03	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	0	55	74	263	392	1,056	13,699	0.042999	0.073363	0.028725	0.039154	0.009119	0.001858	0.957001	0.926637	0.971275	0.960846	0.990881	0.998142	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	2	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	2	20
BTUMQA225K_307ae105f731dc	train	52	00000_052	52	Should edema and necrotic or non-enhancing tumor core be treated as an ambiguous pair in this slice?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_01	ambiguous_subregion_pair_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	81	62	396	539	1,096	13,953	0.033497	0.062273	0.014663	0.02297	0.010359	0.001764	0.966503	0.937727	0.985337	0.97703	0.989641	0.998236	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	3	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	3	20
BTUMQA225K_37df8b26068573	train	52	00000_052	52	Is the reliability difference between enhancing tumor and the peritumoral context region close, moderate, or wide?	moderate_gap	classification	reliability_gap_bucket	reliability_gap_bucket_template_02	reliability_gap_bucket_paraphrase_02	btumqa_225k	bucketed	medium	no	enhancing	context	moderate_gap	reliability_gap_bucket_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	81	62	396	539	1,096	13,953	0.033497	0.062273	0.014663	0.02297	0.010359	0.001764	0.966503	0.937727	0.985337	0.97703	0.989641	0.998236	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	3	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	3	20
BTUMQA225K_87165d0f2b110d	train	52	00000_052	52	How large is the uncertainty gap between enhancing tumor and the peritumoral context region?	moderate_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_01	uncertainty_gap_bucket_paraphrase_01	btumqa_225k	bucketed	medium	no	enhancing	context	moderate_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	81	62	396	539	1,096	13,953	0.033497	0.062273	0.014663	0.02297	0.010359	0.001764	0.966503	0.937727	0.985337	0.97703	0.989641	0.998236	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	3	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	3	20
BTUMQA225K_0b60c5304f6152	train	53	00000_053	53	What best describes the uncertainty separation between enhancing tumor and the peritumoral context region?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_03	uncertainty_gap_bucket_paraphrase_03	btumqa_225k	bucketed	hard	yes	enhancing	context	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	107	54	542	703	1,314	14,200	0.035751	0.0524	0.012709	0.019265	0.012336	0.002166	0.964249	0.9476	0.987291	0.980735	0.987664	0.997834	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	4	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	4	20
BTUMQA225K_390b449db9c0c3	train	53	00000_053	53	What is the confidence-qualified extent of the whole tumor in this slice?	small_uncertain	classification	confidence_qualified_extent	confidence_qualified_extent_template_01	confidence_qualified_extent_paraphrase_01	btumqa_225k	confidence_qualified	hard	no	tumor	null	moderate_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	107	54	542	703	1,314	14,200	0.035751	0.0524	0.012709	0.019265	0.012336	0.002166	0.964249	0.9476	0.987291	0.980735	0.987664	0.997834	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	4	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	4	20
BTUMQA225K_a5ce32851b048e	train	53	00000_053	53	Between enhancing tumor and the peritumoral context region, which token is more reliable?	context	classification	more_reliable_region	more_reliable_region_template_03	more_reliable_region_paraphrase_03	btumqa_225k	comparative	hard	yes	enhancing	context	close_gap	more_reliable_region_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	107	54	542	703	1,314	14,200	0.035751	0.0524	0.012709	0.019265	0.012336	0.002166	0.964249	0.9476	0.987291	0.980735	0.987664	0.997834	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	4	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	4	20
BTUMQA225K_5a9e04b4aa656c	train	54	00000_054	54	Which region is the riskiest evidence source in this MRI slice?	ncr_net	classification	highest_risk_region	highest_risk_region_template_03	highest_risk_region_paraphrase_03	btumqa_225k	ranking	medium	no	ncr_net	edema	moderate_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_63a891f4fb13fb	train	54	00000_054	54	Is necrotic or non-enhancing tumor core confidently present, uncertainly present, or not present here?	uncertain_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_02	confidence_qualified_presence_paraphrase_02	btumqa_225k	confidence_qualified	hard	no	ncr_net	null	moderate_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_ad74ae6f9c9d4b	train	54	00000_054	54	Which region should be trusted more here: the whole tumor region or the peritumoral context region?	context	classification	more_reliable_region	more_reliable_region_template_02	more_reliable_region_paraphrase_02	btumqa_225k	comparative	medium	no	tumor	context	moderate_gap	more_reliable_region_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_bb8fde545885bd	train	54	00000_054	54	Does reliability-aware interpretation preserve the raw dominant tumor component?	consistent	classification	uncertainty_consistency_check	uncertainty_consistency_check_template_03	uncertainty_consistency_check_paraphrase_03	btumqa_225k	ambiguity_sensitive	medium	no	enhancing	enhancing	moderate_gap	uncertainty_consistency_check_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_d584392101e70a	train	54	00000_054	54	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_ec951c8b78d4a4	train	54	00000_054	54	Which tumor subregion is strongest after uncertainty-aware adjustment?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_02	dominant_region_under_uncertainty_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	edema	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	137	89	653	879	1,408	14,494	0.0351	0.046778	0.011802	0.018975	0.013052	0.002502	0.9649	0.953222	0.988198	0.981025	0.986948	0.997498	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	5	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	5	20
BTUMQA225K_0542cee8bbfa4d	train	55	00000_055	55	How large is the uncertainty gap between edema and necrotic or non-enhancing tumor core?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_01	uncertainty_gap_bucket_paraphrase_01	btumqa_225k	bucketed	hard	yes	edema	ncr_net	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	165	116	745	1,026	1,483	14,774	0.037249	0.037247	0.011717	0.01871	0.013199	0.002751	0.962751	0.962753	0.988283	0.98129	0.986801	0.997249	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	6	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	6	20
BTUMQA225K_bba5d3fef8cfa4	train	55	00000_055	55	Between edema and necrotic or non-enhancing tumor core, which region is more reliable for reasoning?	ncr_net	classification	more_reliable_region	more_reliable_region_template_01	more_reliable_region_paraphrase_01	btumqa_225k	comparative	hard	yes	edema	ncr_net	close_gap	more_reliable_region_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	165	116	745	1,026	1,483	14,774	0.037249	0.037247	0.011717	0.01871	0.013199	0.002751	0.962751	0.962753	0.988283	0.98129	0.986801	0.997249	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	6	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	6	20
BTUMQA225K_f5f773e281f6ee	train	55	00000_055	55	Which visual region carries the highest reasoning risk in this slice?	edema	classification	highest_risk_region	highest_risk_region_template_01	highest_risk_region_paraphrase_01	btumqa_225k	ranking	hard	yes	edema	ncr_net	close_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	165	116	745	1,026	1,483	14,774	0.037249	0.037247	0.011717	0.01871	0.013199	0.002751	0.962751	0.962753	0.988283	0.98129	0.986801	0.997249	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	6	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	6	20
BTUMQA225K_80e8a73e6d33b0	train	56	00000_056	56	Which visual region carries the highest reasoning risk in this slice?	edema	classification	highest_risk_region	highest_risk_region_template_01	highest_risk_region_paraphrase_01	btumqa_225k	ranking	easy	no	edema	ncr_net	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	167	167	828	1,162	1,528	15,037	0.03772	0.02223	0.011537	0.016837	0.011456	0.002621	0.96228	0.977771	0.988463	0.983163	0.988544	0.997379	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	7	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	7	20
BTUMQA225K_a700c2d97bfb8f	train	56	00000_056	56	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	167	167	828	1,162	1,528	15,037	0.03772	0.02223	0.011537	0.016837	0.011456	0.002621	0.96228	0.977771	0.988463	0.983163	0.988544	0.997379	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	7	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	7	20
BTUMQA225K_b5c358516c69f3	train	56	00000_056	56	Should edema and necrotic or non-enhancing tumor core be treated as an ambiguous pair in this slice?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_01	ambiguous_subregion_pair_paraphrase_01	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	167	167	828	1,162	1,528	15,037	0.03772	0.02223	0.011537	0.016837	0.011456	0.002621	0.96228	0.977771	0.988463	0.983163	0.988544	0.997379	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	7	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	7	20
BTUMQA225K_daf9e1cb724d83	train	56	00000_056	56	Between enhancing tumor and the whole tumor region, which token is more reliable?	enhancing	classification	more_reliable_region	more_reliable_region_template_03	more_reliable_region_paraphrase_03	btumqa_225k	comparative	medium	no	enhancing	tumor	moderate_gap	more_reliable_region_v1	phase2c_weight	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	167	167	828	1,162	1,528	15,037	0.03772	0.02223	0.011537	0.016837	0.011456	0.002621	0.96228	0.977771	0.988463	0.983163	0.988544	0.997379	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	7	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	7	20
BTUMQA225K_db93225a706595	train	56	00000_056	56	Which region is safest to trust for reasoning in this slice?	context	classification	safe_region_for_reasoning	safe_region_for_reasoning_template_01	safe_region_for_reasoning_paraphrase_01	btumqa_225k	ranking	hard	yes	context	tumor	close_gap	safe_region_for_reasoning_v1	phase2a_area+phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	167	167	828	1,162	1,528	15,037	0.03772	0.02223	0.011537	0.016837	0.011456	0.002621	0.96228	0.977771	0.988463	0.983163	0.988544	0.997379	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	7	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	7	20
BTUMQA225K_06e1de598c7c44	train	57	00000_057	57	Is edema confidently present, uncertainly present, or not present here?	uncertain_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_02	confidence_qualified_presence_paraphrase_02	btumqa_225k	confidence_qualified	hard	no	edema	null	moderate_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_7f484219b7a866	train	57	00000_057	57	How large is the uncertainty gap between enhancing tumor and the peritumoral context region?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_01	uncertainty_gap_bucket_paraphrase_01	btumqa_225k	bucketed	hard	yes	enhancing	context	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_90059085ac8e55	train	57	00000_057	57	Which region is least risky to inspect in this slice?	enhancing	classification	lowest_risk_region	lowest_risk_region_template_02	lowest_risk_region_paraphrase_02	btumqa_225k	ranking	medium	no	enhancing	context	moderate_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_b07bbb7e8cd396	train	57	00000_057	57	Between enhancing tumor and the peritumoral context region, which region is more reliable for reasoning?	enhancing	classification	more_reliable_region	more_reliable_region_template_01	more_reliable_region_paraphrase_01	btumqa_225k	comparative	hard	yes	enhancing	context	close_gap	more_reliable_region_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_b16b2e0dc2f13d	train	57	00000_057	57	Is necrotic or non-enhancing tumor core confidently present, uncertainly present, or not present here?	confident_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_02	confidence_qualified_presence_paraphrase_02	btumqa_225k	confidence_qualified	medium	yes	ncr_net	null	close_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_b561ccdd8ff0d6	train	57	00000_057	57	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_ec2cc8bb079c1e	train	57	00000_057	57	Does this slice show an ambiguity-sensitive competition between edema and necrotic or non-enhancing tumor core?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_03	ambiguous_subregion_pair_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	141	188	938	1,267	1,760	15,351	0.034985	0.014566	0.008931	0.012667	0.009844	0.002322	0.965015	0.985434	0.991069	0.987333	0.990156	0.997678	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	8	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	8	20
BTUMQA225K_0c329f136531e1	train	58	00000_058	58	What is the confidence-qualified presence status of necrotic or non-enhancing tumor core in this slice?	confident_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_01	confidence_qualified_presence_paraphrase_01	btumqa_225k	confidence_qualified	medium	yes	ncr_net	null	close_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	149	241	966	1,356	2,057	15,671	0.034493	0.012992	0.006848	0.010977	0.00888	0.002209	0.965507	0.987008	0.993152	0.989023	0.99112	0.997791	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	9	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	9	20
BTUMQA225K_14199b7e9a978b	train	58	00000_058	58	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	149	241	966	1,356	2,057	15,671	0.034493	0.012992	0.006848	0.010977	0.00888	0.002209	0.965507	0.987008	0.993152	0.989023	0.99112	0.997791	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	9	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	9	20
BTUMQA225K_4440fdbfa3e4a0	train	58	00000_058	58	What best describes the reliability separation between necrotic or non-enhancing tumor core and the whole tumor region?	close_gap	classification	reliability_gap_bucket	reliability_gap_bucket_template_03	reliability_gap_bucket_paraphrase_03	btumqa_225k	bucketed	hard	yes	ncr_net	tumor	close_gap	reliability_gap_bucket_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	149	241	966	1,356	2,057	15,671	0.034493	0.012992	0.006848	0.010977	0.00888	0.002209	0.965507	0.987008	0.993152	0.989023	0.99112	0.997791	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	9	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	9	20
BTUMQA225K_83b34a3c99efc0	train	58	00000_058	58	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	149	241	966	1,356	2,057	15,671	0.034493	0.012992	0.006848	0.010977	0.00888	0.002209	0.965507	0.987008	0.993152	0.989023	0.99112	0.997791	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	9	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	9	20
BTUMQA225K_89f6c78f463758	train	58	00000_058	58	What is the confidence-qualified presence status of edema in this slice?	uncertain_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_01	confidence_qualified_presence_paraphrase_01	btumqa_225k	confidence_qualified	hard	no	edema	null	moderate_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	149	241	966	1,356	2,057	15,671	0.034493	0.012992	0.006848	0.010977	0.00888	0.002209	0.965507	0.987008	0.993152	0.989023	0.99112	0.997791	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	9	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	9	20
BTUMQA225K_96421c016ac8eb	train	59	00000_059	59	Is the uncertainty difference between necrotic or non-enhancing tumor core and the peritumoral context region close, moderate, or wide?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_02	uncertainty_gap_bucket_paraphrase_02	btumqa_225k	bucketed	hard	yes	ncr_net	context	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	169	303	956	1,428	2,347	15,996	0.032122	0.008711	0.006566	0.010046	0.008019	0.002242	0.967878	0.991289	0.993434	0.989954	0.991981	0.997758	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	10	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	10	20
BTUMQA225K_9bfc09c88f6e2e	train	59	00000_059	59	What is the confidence-qualified extent of the whole tumor in this slice?	moderate_confident	classification	confidence_qualified_extent	confidence_qualified_extent_template_01	confidence_qualified_extent_paraphrase_01	btumqa_225k	confidence_qualified	medium	yes	tumor	null	close_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	169	303	956	1,428	2,347	15,996	0.032122	0.008711	0.006566	0.010046	0.008019	0.002242	0.967878	0.991289	0.993434	0.989954	0.991981	0.997758	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	10	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	10	20
BTUMQA225K_c456e6a3d6c298	train	59	00000_059	59	Which tumor subregion is strongest after uncertainty-aware adjustment?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_02	dominant_region_under_uncertainty_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	169	303	956	1,428	2,347	15,996	0.032122	0.008711	0.006566	0.010046	0.008019	0.002242	0.967878	0.991289	0.993434	0.989954	0.991981	0.997758	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	10	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	10	20
BTUMQA225K_e59ddc441d41c9	train	59	00000_059	59	Which region should be trusted more here: necrotic or non-enhancing tumor core or the peritumoral context region?	context	classification	more_reliable_region	more_reliable_region_template_02	more_reliable_region_paraphrase_02	btumqa_225k	comparative	hard	yes	ncr_net	context	close_gap	more_reliable_region_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	169	303	956	1,428	2,347	15,996	0.032122	0.008711	0.006566	0.010046	0.008019	0.002242	0.967878	0.991289	0.993434	0.989954	0.991981	0.997758	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	10	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	10	20
BTUMQA225K_e724c36cbeeeb4	train	59	00000_059	59	Between necrotic or non-enhancing tumor core and the whole tumor region, where is the uncertainty higher?	tumor	classification	more_uncertain_region	more_uncertain_region_template_03	more_uncertain_region_paraphrase_03	btumqa_225k	comparative	hard	yes	ncr_net	tumor	close_gap	more_uncertain_region_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	169	303	956	1,428	2,347	15,996	0.032122	0.008711	0.006566	0.010046	0.008019	0.002242	0.967878	0.991289	0.993434	0.989954	0.991981	0.997758	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	10	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	10	20
BTUMQA225K_0f3ae0ea118cb0	train	60	00000_060	60	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_22a99d381b0ed7	train	60	00000_060	60	Which region is least risky to inspect in this slice?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_02	lowest_risk_region_paraphrase_02	btumqa_225k	ranking	hard	yes	ncr_net	enhancing	close_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_2b2a43cdcb247f	train	60	00000_060	60	Which confidence-qualified extent label best fits the tumor region here?	moderate_confident	classification	confidence_qualified_extent	confidence_qualified_extent_template_03	confidence_qualified_extent_paraphrase_03	btumqa_225k	confidence_qualified	medium	yes	tumor	null	close_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_2bd15ecfcd2068	train	60	00000_060	60	What is the confidence-qualified presence status of edema in this slice?	uncertain_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_01	confidence_qualified_presence_paraphrase_01	btumqa_225k	confidence_qualified	hard	no	edema	null	moderate_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_30a563929fa89b	train	60	00000_060	60	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	edema	ncr_net	wide_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_5f296731da13b4	train	60	00000_060	60	Does uncertainty-aware dominance agree with raw area dominance in this slice?	consistent	classification	uncertainty_consistency_check	uncertainty_consistency_check_template_01	uncertainty_consistency_check_paraphrase_01	btumqa_225k	ambiguity_sensitive	medium	no	enhancing	enhancing	moderate_gap	uncertainty_consistency_check_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_d9c8a4024f62af	train	60	00000_060	60	Which visual region carries the highest reasoning risk in this slice?	edema	classification	highest_risk_region	highest_risk_region_template_01	highest_risk_region_paraphrase_01	btumqa_225k	ranking	easy	no	edema	tumor	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	207	341	945	1,493	2,092	16,277	0.030391	0.00592	0.005396	0.008981	0.007743	0.002148	0.969609	0.99408	0.994604	0.991019	0.992257	0.997852	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	11	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	11	20
BTUMQA225K_0a3901a6429de5	train	61	00000_061	61	Between necrotic or non-enhancing tumor core and enhancing tumor, which region is more reliable for reasoning?	enhancing	classification	more_reliable_region	more_reliable_region_template_01	more_reliable_region_paraphrase_01	btumqa_225k	comparative	hard	yes	ncr_net	enhancing	close_gap	more_reliable_region_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	229	401	946	1,576	2,354	16,520	0.026739	0.006695	0.005718	0.009021	0.007097	0.002158	0.973261	0.993305	0.994282	0.990979	0.992903	0.997842	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	12	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	12	20
BTUMQA225K_2ca2acd7574e8b	train	61	00000_061	61	Which visual region carries the lowest reasoning risk in this slice?	enhancing	classification	lowest_risk_region	lowest_risk_region_template_01	lowest_risk_region_paraphrase_01	btumqa_225k	ranking	hard	yes	enhancing	ncr_net	close_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	229	401	946	1,576	2,354	16,520	0.026739	0.006695	0.005718	0.009021	0.007097	0.002158	0.973261	0.993305	0.994282	0.990979	0.992903	0.997842	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	12	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	12	20
BTUMQA225K_3a6c3a24f9dc7d	train	61	00000_061	61	Is the uncertainty difference between necrotic or non-enhancing tumor core and the peritumoral context region close, moderate, or wide?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_02	uncertainty_gap_bucket_paraphrase_02	btumqa_225k	bucketed	hard	yes	ncr_net	context	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	229	401	946	1,576	2,354	16,520	0.026739	0.006695	0.005718	0.009021	0.007097	0.002158	0.973261	0.993305	0.994282	0.990979	0.992903	0.997842	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	12	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	12	20
BTUMQA225K_58d9947f304840	train	61	00000_061	61	Which confidence-qualified extent label best fits the tumor region here?	moderate_confident	classification	confidence_qualified_extent	confidence_qualified_extent_template_03	confidence_qualified_extent_paraphrase_03	btumqa_225k	confidence_qualified	medium	yes	tumor	null	close_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	229	401	946	1,576	2,354	16,520	0.026739	0.006695	0.005718	0.009021	0.007097	0.002158	0.973261	0.993305	0.994282	0.990979	0.992903	0.997842	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	12	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	12	20
BTUMQA225K_d8a7afa823c2d9	train	61	00000_061	61	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	229	401	946	1,576	2,354	16,520	0.026739	0.006695	0.005718	0.009021	0.007097	0.002158	0.973261	0.993305	0.994282	0.990979	0.992903	0.997842	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	12	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	12	20
BTUMQA225K_c35bfd3d8d6c10	train	62	00000_062	62	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	313	489	904	1,706	2,653	16,721	0.022111	0.005195	0.005713	0.008573	0.007026	0.002154	0.977889	0.994805	0.994287	0.991427	0.992974	0.997846	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	13	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	13	20
BTUMQA225K_52a634821fc95b	train	63	00000_063	63	Which region has the lowest evidence risk here?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_03	lowest_risk_region_paraphrase_03	btumqa_225k	ranking	medium	no	ncr_net	enhancing	moderate_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_8615257e86f5ee	train	63	00000_063	63	Between necrotic or non-enhancing tumor core and enhancing tumor, where is the uncertainty higher?	enhancing	classification	more_uncertain_region	more_uncertain_region_template_03	more_uncertain_region_paraphrase_03	btumqa_225k	comparative	hard	yes	ncr_net	enhancing	close_gap	more_uncertain_region_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_9391f78e6020fc	train	63	00000_063	63	How should the presence of enhancing tumor be described in this slice?	confident_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_03	confidence_qualified_presence_paraphrase_03	btumqa_225k	confidence_qualified	medium	yes	enhancing	null	close_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_a6e687b795037c	train	63	00000_063	63	What is the confidence-qualified extent of the whole tumor in this slice?	moderate_confident	classification	confidence_qualified_extent	confidence_qualified_extent_template_01	confidence_qualified_extent_paraphrase_01	btumqa_225k	confidence_qualified	medium	yes	tumor	null	close_gap	confidence_qualified_extent_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_ba667a9f079ee2	train	63	00000_063	63	Which region should be treated with the most caution for downstream reasoning?	edema	classification	highest_risk_region	highest_risk_region_template_02	highest_risk_region_paraphrase_02	btumqa_225k	ranking	easy	no	edema	tumor	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_e1c4247b273805	train	63	00000_063	63	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	453	558	876	1,887	2,481	16,881	0.018964	0.004941	0.005372	0.008507	0.007293	0.002186	0.981036	0.995059	0.994628	0.991493	0.992707	0.997814	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	14	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	14	20
BTUMQA225K_1db8a60e46614c	train	64	00000_064	64	Is the uncertainty difference between the whole tumor region and the peritumoral context region close, moderate, or wide?	close_gap	classification	uncertainty_gap_bucket	uncertainty_gap_bucket_template_02	uncertainty_gap_bucket_paraphrase_02	btumqa_225k	bucketed	hard	yes	tumor	context	close_gap	uncertainty_gap_bucket_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	507	556	932	1,995	2,190	17,047	0.016516	0.003459	0.005287	0.007631	0.00815	0.002213	0.983484	0.996541	0.994713	0.992369	0.99185	0.997787	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	15	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	15	20
BTUMQA225K_26178bd7eea4e5	train	64	00000_064	64	Which region should receive the highest trust for reasoning?	context	classification	safe_region_for_reasoning	safe_region_for_reasoning_template_03	safe_region_for_reasoning_paraphrase_03	btumqa_225k	ranking	hard	yes	context	tumor	close_gap	safe_region_for_reasoning_v1	phase2a_area+phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	507	556	932	1,995	2,190	17,047	0.016516	0.003459	0.005287	0.007631	0.00815	0.002213	0.983484	0.996541	0.994713	0.992369	0.99185	0.997787	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	15	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	15	20
BTUMQA225K_368fc61ddf3a78	train	64	00000_064	64	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	507	556	932	1,995	2,190	17,047	0.016516	0.003459	0.005287	0.007631	0.00815	0.002213	0.983484	0.996541	0.994713	0.992369	0.99185	0.997787	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	15	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	15	20
BTUMQA225K_5297a222fe71bd	train	64	00000_064	64	How large is the reliability gap between the whole tumor region and the peritumoral context region?	close_gap	classification	reliability_gap_bucket	reliability_gap_bucket_template_01	reliability_gap_bucket_paraphrase_01	btumqa_225k	bucketed	hard	yes	tumor	context	close_gap	reliability_gap_bucket_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	507	556	932	1,995	2,190	17,047	0.016516	0.003459	0.005287	0.007631	0.00815	0.002213	0.983484	0.996541	0.994713	0.992369	0.99185	0.997787	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	15	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	15	20
BTUMQA225K_aa49b1ecb40432	train	64	00000_064	64	Between the whole tumor region and the peritumoral context region, which region is more uncertain in this slice?	context	classification	more_uncertain_region	more_uncertain_region_template_01	more_uncertain_region_paraphrase_01	btumqa_225k	comparative	hard	yes	tumor	context	close_gap	more_uncertain_region_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	507	556	932	1,995	2,190	17,047	0.016516	0.003459	0.005287	0.007631	0.00815	0.002213	0.983484	0.996541	0.994713	0.992369	0.99185	0.997787	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	15	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	15	20
BTUMQA225K_0b710f1ac92e7e	train	65	00000_065	65	Which visual region carries the lowest reasoning risk in this slice?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_01	lowest_risk_region_paraphrase_01	btumqa_225k	ranking	medium	no	ncr_net	enhancing	moderate_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_18b904a5eaadc6	train	65	00000_065	65	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_721dcc7bba1bd6	train	65	00000_065	65	How should the presence of edema be described in this slice?	uncertain_present	classification	confidence_qualified_presence	confidence_qualified_presence_template_03	confidence_qualified_presence_paraphrase_03	btumqa_225k	confidence_qualified	hard	no	edema	null	moderate_gap	confidence_qualified_presence_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_baa970ca9724c5	train	65	00000_065	65	What best describes the reliability separation between necrotic or non-enhancing tumor core and enhancing tumor?	close_gap	classification	reliability_gap_bucket	reliability_gap_bucket_template_03	reliability_gap_bucket_paraphrase_03	btumqa_225k	bucketed	hard	yes	ncr_net	enhancing	close_gap	reliability_gap_bucket_v1	phase2c_weight	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_e18299b4e6e055	train	65	00000_065	65	Which tumor subregion is strongest after uncertainty-aware adjustment?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_02	dominant_region_under_uncertainty_paraphrase_02	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_f14e881559b593	train	65	00000_065	65	Which region should be treated with the most caution for downstream reasoning?	edema	classification	highest_risk_region	highest_risk_region_template_02	highest_risk_region_paraphrase_02	btumqa_225k	ranking	easy	no	edema	context	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	538	554	971	2,063	2,290	17,191	0.018	0.003349	0.004152	0.007548	0.009454	0.002373	0.982	0.996651	0.995848	0.992452	0.990546	0.997627	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	16	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	16	20
BTUMQA225K_5cbd13e44b6e2b	train	66	00000_066	66	Which visual region carries the highest reasoning risk in this slice?	edema	classification	highest_risk_region	highest_risk_region_template_01	highest_risk_region_paraphrase_01	btumqa_225k	ranking	easy	no	edema	context	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	608	581	966	2,155	2,536	17,330	0.01897	0.004713	0.005103	0.00891	0.008835	0.002565	0.98103	0.995287	0.994897	0.99109	0.991165	0.997435	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	17	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	17	20
BTUMQA225K_7390a41526fbfc	train	66	00000_066	66	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	608	581	966	2,155	2,536	17,330	0.01897	0.004713	0.005103	0.00891	0.008835	0.002565	0.98103	0.995287	0.994897	0.99109	0.991165	0.997435	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	17	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	17	20
BTUMQA225K_99eafb6e754229	train	66	00000_066	66	Which visual region carries the lowest reasoning risk in this slice?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_01	lowest_risk_region_paraphrase_01	btumqa_225k	ranking	medium	no	ncr_net	enhancing	moderate_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	608	581	966	2,155	2,536	17,330	0.01897	0.004713	0.005103	0.00891	0.008835	0.002565	0.98103	0.995287	0.994897	0.99109	0.991165	0.997435	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	17	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	17	20
BTUMQA225K_e141243238373b	train	66	00000_066	66	Does this slice show an ambiguity-sensitive competition between edema and necrotic or non-enhancing tumor core?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_03	ambiguous_subregion_pair_paraphrase_03	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	608	581	966	2,155	2,536	17,330	0.01897	0.004713	0.005103	0.00891	0.008835	0.002565	0.98103	0.995287	0.994897	0.99109	0.991165	0.997435	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	17	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	17	20
BTUMQA225K_03eedae0dd1e33	train	67	00000_067	67	After accounting for reliability, which tumor subregion remains most dominant in this slice?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_01	dominant_region_under_uncertainty_paraphrase_01	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	684	594	938	2,216	2,513	17,418	0.019826	0.004726	0.004473	0.00928	0.009172	0.00265	0.980174	0.995274	0.995527	0.99072	0.990828	0.99735	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	18	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	18	20
BTUMQA225K_10206e8cdcfc21	train	67	00000_067	67	Should edema and necrotic or non-enhancing tumor core be treated as an ambiguous pair in this slice?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_01	ambiguous_subregion_pair_paraphrase_01	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	684	594	938	2,216	2,513	17,418	0.019826	0.004726	0.004473	0.00928	0.009172	0.00265	0.980174	0.995274	0.995527	0.99072	0.990828	0.99735	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	18	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	18	20
BTUMQA225K_2b6484cdfb359b	train	67	00000_067	67	Which visual region carries the lowest reasoning risk in this slice?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_01	lowest_risk_region_paraphrase_01	btumqa_225k	ranking	hard	yes	ncr_net	enhancing	close_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	684	594	938	2,216	2,513	17,418	0.019826	0.004726	0.004473	0.00928	0.009172	0.00265	0.980174	0.995274	0.995527	0.99072	0.990828	0.99735	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	18	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	18	20
BTUMQA225K_6bd0ad4b133459	train	67	00000_067	67	Which visual region carries the highest reasoning risk in this slice?	edema	classification	highest_risk_region	highest_risk_region_template_01	highest_risk_region_paraphrase_01	btumqa_225k	ranking	easy	no	edema	context	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	684	594	938	2,216	2,513	17,418	0.019826	0.004726	0.004473	0.00928	0.009172	0.00265	0.980174	0.995274	0.995527	0.99072	0.990828	0.99735	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	18	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	18	20
BTUMQA225K_16c1ad788351ad	train	68	00000_068	68	Should edema and necrotic or non-enhancing tumor core be treated as an ambiguous pair in this slice?	ambiguous	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_01	ambiguous_subregion_pair_paraphrase_01	btumqa_225k	ambiguity_sensitive	hard	yes	edema	ncr_net	close_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	753	578	948	2,279	2,499	17,476	0.018879	0.004603	0.004635	0.009333	0.009944	0.002775	0.981121	0.995397	0.995365	0.990667	0.990056	0.997225	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	19	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	19	20
BTUMQA225K_52a39d3aaf3144	train	68	00000_068	68	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	ncr_net	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	753	578	948	2,279	2,499	17,476	0.018879	0.004603	0.004635	0.009333	0.009944	0.002775	0.981121	0.995397	0.995365	0.990667	0.990056	0.997225	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	19	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	19	20
BTUMQA225K_5595bff1dcc651	train	68	00000_068	68	Which region has the lowest evidence risk here?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_03	lowest_risk_region_paraphrase_03	btumqa_225k	ranking	hard	yes	ncr_net	enhancing	close_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	753	578	948	2,279	2,499	17,476	0.018879	0.004603	0.004635	0.009333	0.009944	0.002775	0.981121	0.995397	0.995365	0.990667	0.990056	0.997225	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	19	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	19	20
BTUMQA225K_e7c1163b47f074	train	68	00000_068	68	Does reliability-aware interpretation preserve the raw dominant tumor component?	consistent	classification	uncertainty_consistency_check	uncertainty_consistency_check_template_03	uncertainty_consistency_check_paraphrase_03	btumqa_225k	ambiguity_sensitive	medium	no	enhancing	enhancing	moderate_gap	uncertainty_consistency_check_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	753	578	948	2,279	2,499	17,476	0.018879	0.004603	0.004635	0.009333	0.009944	0.002775	0.981121	0.995397	0.995365	0.990667	0.990056	0.997225	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	19	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	19	20
BTUMQA225K_ea6a904ab4bea9	train	68	00000_068	68	Which region should be treated with the most caution for downstream reasoning?	edema	classification	highest_risk_region	highest_risk_region_template_02	highest_risk_region_paraphrase_02	btumqa_225k	ranking	easy	no	edema	context	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	753	578	948	2,279	2,499	17,476	0.018879	0.004603	0.004635	0.009333	0.009944	0.002775	0.981121	0.995397	0.995365	0.990667	0.990056	0.997225	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	19	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	19	20
BTUMQA225K_143f1cb022aa0b	train	69	00000_069	69	Between necrotic or non-enhancing tumor core and enhancing tumor, which region is more uncertain in this slice?	ncr_net	classification	more_uncertain_region	more_uncertain_region_template_01	more_uncertain_region_paraphrase_01	btumqa_225k	comparative	hard	yes	ncr_net	enhancing	close_gap	more_uncertain_region_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_677b62298c7f69	train	69	00000_069	69	Which region has less reliable visual evidence here: the whole tumor region or the peritumoral context region?	context	classification	more_uncertain_region	more_uncertain_region_template_02	more_uncertain_region_paraphrase_02	btumqa_225k	comparative	hard	yes	tumor	context	close_gap	more_uncertain_region_v1	phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_75cc7e34d993e3	train	69	00000_069	69	Which region is the riskiest evidence source in this MRI slice?	edema	classification	highest_risk_region	highest_risk_region_template_03	highest_risk_region_paraphrase_03	btumqa_225k	ranking	easy	no	edema	context	wide_gap	highest_risk_region_v1	phase2a_area+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_76953e955324ad	train	69	00000_069	69	Which region has the lowest evidence risk here?	ncr_net	classification	lowest_risk_region	lowest_risk_region_template_03	lowest_risk_region_paraphrase_03	btumqa_225k	ranking	hard	yes	ncr_net	enhancing	close_gap	lowest_risk_region_v1	phase2a_area+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_8389872aea93ba	train	69	00000_069	69	Are edema and necrotic or non-enhancing tumor core too close to separate confidently here?	ambiguous	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_02	ambiguous_subregion_pair_paraphrase_02	btumqa_225k	ambiguity_sensitive	hard	yes	edema	ncr_net	close_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	ambiguity_priority	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_b0e8b827d1ed75	train	69	00000_069	69	Under uncertainty-aware interpretation, which tumor subregion is dominant here?	enhancing	classification	dominant_region_under_uncertainty	dominant_region_under_uncertainty_template_03	dominant_region_under_uncertainty_paraphrase_03	btumqa_225k	ambiguity_sensitive	easy	no	enhancing	edema	wide_gap	dominant_region_under_uncertainty_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	799	572	954	2,325	2,523	17,544	0.018628	0.004721	0.004654	0.009473	0.010842	0.002946	0.981372	0.995279	0.995346	0.990527	0.989158	0.997054	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	20	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	20	20
BTUMQA225K_1c9e460d2ba32d	train	70	00000_070	70	Does this slice show an ambiguity-sensitive competition between edema and necrotic or non-enhancing tumor core?	distinct	classification	ambiguous_subregion_pair	ambiguous_subregion_pair_template_03	ambiguous_subregion_pair_paraphrase_03	btumqa_225k	ambiguity_sensitive	medium	no	edema	ncr_net	moderate_gap	ambiguous_subregion_pair_v1	phase1_pseudo_mask+phase2b_uncertainty	balanced_family_sampling	FLAIR\|T1\|T1ce\|T2	1	879	558	964	2,401	2,525	17,579	0.019536	0.003797	0.00448	0.009833	0.011157	0.003087	0.980464	0.996203	0.99552	0.990167	0.988843	0.996913	phase_2/p2a_region_tokens/tokens_by_patient/00000.pt	21	phase_2/p2c_ugtm_modulated_tokens/modulated_tokens_by_patient/00000.pt	21	20

End of preview. Expand in Data Studio

BTUMQA: Brain Tumor MRI Visual Question Answering Dataset

BTUMQA is a large-scale visual question answering dataset designed for controlled research evaluation of brain tumor MRI scans. It contains 225,000 QA pairs mapped to 2D MRI slices, derived from the RSNA BraTS 2021 dataset.

This is the official dataset repository for the paper: "AUGR-VQA: Question-Adaptive Uncertainty-Guided Regional Reasoning for Reliable Brain Tumor MRI Visual Question Answering".

🔗 Paper Code Repository: shuvomonowar00/AUGR-VQA (Private, publishing soon)

This dataset is a controlled research evaluation resource and is not intended for clinical diagnosis, patient treatment, or deployment in real clinical workflows.

📊 Benchmark Readiness & Clean Metadata Protocol

BTUMQA has passed rigorous data integrity and split-leakage checks, making it a robust benchmark for medical VQA.

To guarantee scientific rigor, the dataset features a strict "Clean Metadata Protocol" that separates valid model inputs from diagnostic metadata.

🛑 Warning: Shortcut Learning Risks

During our Phase 03C audit, we found that certain diagnostic fields contain high mutual information with the target answer.

Do not use the following fields as inputs to your VQA model during training. Doing so will cause your model to "cheat" via metadata shortcuts rather than learning visual reasoning.

candidate_keep_reason (Cramer's V correlation = 0.79)

decision_rule_id (Cramer's V correlation = 0.70)

label_provenance (Cramer's V correlation = 0.57)

These fields are provided strictly for post-hoc auditing and error analysis.

Disclaimer: This dataset does not claim to be perfectly bias-free. While the clean protocol drastically reduces direct metadata shortcut risks, residual biases (such as question-template or answer-prior drift) remain measurable and should be actively reported when evaluating models.

Dataset Analytics

(Diagnostic visuals from the Phase 03C Clean Metadata Audit)

Answer Distribution	Shortcut Risk Associations

📂 Dataset Structure

The data files are located in the btumqa_225k/ folder:

btumqa_225k_qa_pairs.csv (225,000 pairs - full pool)
btumqa_225k_train.csv (157,501 pairs)
btumqa_225k_val.csv (33,750 pairs)
btumqa_225k_test.csv (33,750 pairs)

Data Fields (Categorized by Protocol)

✅ Allowed Clean Model Inputs

question: The natural language question regarding the MRI slice.
answer: The target ground truth answer.
modality_context: Available MRI modalities (e.g., FLAIR|T1|T1ce|T2).
phase2a_token_file: Path to the extracted visual tokens.
phase2c_file: Path to the UGTM-modulated tokens.
(Note: Token files are dynamically generated via the official AUGR-VQA code repository)

🛡️ Clean Safe Audit Descriptors

qa_id, patient_id, slice_id, unique_id: Unique identifiers.
split: Dataset split (train, val, test).
difficulty_level: Question difficulty (easy, medium, hard).
question_family & question_style: Semantic templates and formatting styles.
region_target_primary & region_target_secondary: Target brain regions referred to.
ambiguity_flag & signal_gap_bucket: Ambiguity descriptors.

🛑 Forbidden Audit-Only Fields (Do Not Train)

Area Statistics: edema_area_pixels, ncr_net_area_pixels, enhancing_area_pixels, tumor_area_pixels, context_area_pixels, global_area_pixels.
Uncertainty Estimates: edema_uncertainty, ncr_net_uncertainty, enhancing_uncertainty, tumor_uncertainty, context_uncertainty, global_uncertainty.
Modulation Weights: edema_weight, ncr_net_weight, enhancing_weight, tumor_weight, context_weight, global_weight.
Pipeline Provenance: decision_rule_id, label_provenance, candidate_keep_reason, mc_dropout_passes.

🚀 Quick Start

Load the dataset seamlessly using the Hugging Face datasets library:

from datasets import load_dataset

# Load the training split
train_dataset = load_dataset("shuvomonowar00/BTUMQA", data_files="btumqa_225k/btumqa_225k_train.csv")

# Load the validation split
val_dataset = load_dataset("shuvomonowar00/BTUMQA", data_files="btumqa_225k/btumqa_225k_val.csv")

# Load the testing split
test_dataset = load_dataset("shuvomonowar00/BTUMQA", data_files="btumqa_225k/btumqa_225k_test.csv")

Downloads last month: 180