{"id":"https://openalex.org/W7208710503","doi":"https://doi.org/10.48550/arxiv.2609.03085","title":"Solving the Needle-in-a-Haystack Problem in Mammography Vision-Language Model with Differentiable Subset Sampling","display_name":"Solving the Needle-in-a-Haystack Problem in Mammography Vision-Language Model with Differentiable Subset Sampling","publication_year":2026,"publication_date":"2026-09-02","ids":{"openalex":"https://openalex.org/W7208710503","doi":"https://doi.org/10.48550/arxiv.2609.03085"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.03085","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.03085","pdf_url":"https://arxiv.org/pdf/2609.03085","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.03085","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5031359624","display_name":"Young Seok Jeon","orcid":"https://orcid.org/0000-0002-4948-083X","observed_orcids":["https://orcid.org/0000-0002-4948-083X"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jeon, Young Seok","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092187767","display_name":"Beatrice Brown-Mulry","orcid":"https://orcid.org/0009-0003-2835-3852","observed_orcids":["https://orcid.org/0009-0003-2835-3852"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Brown-Mulry, Beatrice","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047273337","display_name":"Rohan Isaac","orcid":"https://orcid.org/0009-0002-7261-678X","observed_orcids":["https://orcid.org/0009-0002-7261-678X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Isaac, Rohan Satya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023464840","display_name":"Anjana Dissanayaka","orcid":"https://orcid.org/0000-0002-9568-5545","observed_orcids":["https://orcid.org/0000-0002-9568-5545"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dissanayaka, Anjana","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037108911","display_name":"Theo Dapamede","orcid":"https://orcid.org/0000-0002-6264-2282","observed_orcids":["https://orcid.org/0000-0002-6264-2282"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dapamede, Theo","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5009060599","display_name":"Mohammadreza Chavoshi","orcid":"https://orcid.org/0000-0001-6598-7190","observed_orcids":["https://orcid.org/0000-0001-6598-7190"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chavoshi, Mohammadreza","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076075666","display_name":"Judy Wawira Gichoya","orcid":"https://orcid.org/0000-0002-1097-316X","observed_orcids":["https://orcid.org/0000-0002-1097-316X"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gichoya, Judy","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5050086721","display_name":"Hari Trivedi","orcid":"https://orcid.org/0000-0001-6648-8334","observed_orcids":["https://orcid.org/0000-0001-6648-8334"]},"institutions":[{"id":"https://openalex.org/I150468666","display_name":"Emory University","ror":"https://ror.org/03czfpz43","country_code":"US","type":"education","lineage":["https://openalex.org/I150468666"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Trivedi, Hari","raw_affiliation_strings":["Department of Radiology , Emory University , Atlanta , GA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Radiology , Emory University , Atlanta , GA , USA","institution_ids":["https://openalex.org/I150468666"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10862","display_name":"AI in cancer detection","score":0.5770999789237976,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10862","display_name":"AI in cancer detection","score":0.5770999789237976,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.35249999165534973,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12422","display_name":"Radiomics and Machine Learning in Medical Imaging","score":0.023000000044703484,"subfield":{"id":"https://openalex.org/subfields/2741","display_name":"Radiology, Nuclear Medicine and Imaging"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/mammography","display_name":"mammography","score":0.8870000243186951},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.9940000176429749},{"id":"https://openalex.org/keywords/clip","display_name":"CLIP","score":0.8199999928474426},{"id":"https://openalex.org/keywords/lesion-localization","display_name":"lesion localization","score":0.8399999737739563},{"id":"https://openalex.org/keywords/bi-rads-classification","display_name":"BI-RADS classification","score":0.7620000243186951},{"id":"https://openalex.org/keywords/breast-cancer-prediction","display_name":"breast cancer prediction","score":0.796999990940094},{"id":"https://openalex.org/keywords/structured-radiology-reports","display_name":"structured radiology reports","score":0.5590000152587891},{"id":"https://openalex.org/keywords/zero-shot-learning","display_name":"zero-shot learning","score":0.6690000295639038},{"id":"https://openalex.org/keywords/linear-probing","display_name":"linear probing","score":0.7160000205039978}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7346000075340271},{"id":"https://openalex.org/C2780472235","wikidata":"https://www.wikidata.org/wiki/Q324634","display_name":"Mammography","level":4,"score":0.7027000188827515},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5982999801635742},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.5091999769210815},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.4722000062465668},{"id":"https://openalex.org/C81388566","wikidata":"https://www.wikidata.org/wiki/Q526668","display_name":"Sigmoid function","level":3,"score":0.45410001277923584},{"id":"https://openalex.org/C202615002","wikidata":"https://www.wikidata.org/wiki/Q783507","display_name":"Differentiable function","level":2,"score":0.44850000739097595},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.4323999881744385},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3968000113964081},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3725999891757965},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.34850001335144043},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C197129107","wikidata":"https://www.wikidata.org/wiki/Q1921621","display_name":"Merge (version control)","level":2,"score":0.3386000096797943},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.3305000066757202},{"id":"https://openalex.org/C31601959","wikidata":"https://www.wikidata.org/wiki/Q931309","display_name":"Medical imaging","level":2,"score":0.32910001277923584},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.32260000705718994},{"id":"https://openalex.org/C142259097","wikidata":"https://www.wikidata.org/wiki/Q5891314","display_name":"Homogeneity (statistics)","level":2,"score":0.27799999713897705},{"id":"https://openalex.org/C75294576","wikidata":"https://www.wikidata.org/wiki/Q5165192","display_name":"Contextual image classification","level":3,"score":0.27489998936653137},{"id":"https://openalex.org/C58471807","wikidata":"https://www.wikidata.org/wiki/Q327120","display_name":"Receiver operating characteristic","level":2,"score":0.27250000834465027},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2689000070095062},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.2646999955177307},{"id":"https://openalex.org/C534262118","wikidata":"https://www.wikidata.org/wiki/Q177719","display_name":"Medical diagnosis","level":2,"score":0.2587999999523163}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.03085","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.03085","pdf_url":"https://arxiv.org/pdf/2609.03085","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.03085","doi":"https://doi.org/10.48550/arxiv.2609.03085","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.03085","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.03085","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.03085","pdf_url":"https://arxiv.org/pdf/2609.03085","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"score":0.46810001134872437,"id":"https://metadata.un.org/sdg/3","display_name":"Good health and well-being"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7208710503.pdf","grobid_xml":"https://content.openalex.org/works/W7208710503.grobid-xml"},"referenced_works_count":38,"referenced_works":["https://openalex.org/W639708223","https://openalex.org/W1959608418","https://openalex.org/W2530010084","https://openalex.org/W2547875792","https://openalex.org/W2548228487","https://openalex.org/W2776207810","https://openalex.org/W2908510526","https://openalex.org/W2946948417","https://openalex.org/W2947798871","https://openalex.org/W3005680577","https://openalex.org/W3035524453","https://openalex.org/W3091546937","https://openalex.org/W3091940082","https://openalex.org/W3094502228","https://openalex.org/W3135367836","https://openalex.org/W3206072662","https://openalex.org/W3213454282","https://openalex.org/W4220972931","https://openalex.org/W4297808394","https://openalex.org/W4306820534","https://openalex.org/W4312881242","https://openalex.org/W4312933868","https://openalex.org/W4312980231","https://openalex.org/W4323066559","https://openalex.org/W4366330503","https://openalex.org/W4385573131","https://openalex.org/W4401306886","https://openalex.org/W4403345210","https://openalex.org/W4403649762","https://openalex.org/W4404988052","https://openalex.org/W4407025656","https://openalex.org/W4407815020","https://openalex.org/W4413003144","https://openalex.org/W4414687708","https://openalex.org/W4415795821","https://openalex.org/W4416932938","https://openalex.org/W7123459016","https://openalex.org/W7133197087"],"related_works":["https://openalex.org/W7125417563","https://openalex.org/W7161798660","https://openalex.org/W4410696141","https://openalex.org/W7169816981","https://openalex.org/W4413003144","https://openalex.org/W4403649762","https://openalex.org/W4403649890","https://openalex.org/W7161792751","https://openalex.org/W4410297876","https://openalex.org/W4401749569"],"abstract_inverted_index":{"There":[0],"is":[1,46],"growing":[2],"interest":[3],"in":[4,29],"adopting":[5],"CLIP-style":[6],"vision--language":[7],"model":[8],"(VLM)":[9],"pretraining":[10],"for":[11],"mammography.":[12],"However,":[13],"models":[14],"that":[15,42,108],"directly":[16],"employ":[17],"the":[18,123,147,163],"standard":[19],"CLIP":[20],"architecture":[21,89],"and":[22,37,59,90,180,187,197,216,234],"training":[23,218],"objective":[24],"exhibit":[25],"limited":[26],"zero-shot":[27,201],"performance":[28,45],"clinically":[30],"important":[31],"tasks":[32],"such":[33],"as":[34,136],"cancer,":[35],"finding-type,":[36],"BI-RADS":[38,193],"predictions.":[39],"We":[40,75],"argue":[41],"this":[43],"underwhelming":[44],"due":[47],"to":[48,81,99,110,119],"neglecting":[49],"two":[50,84],"characteristics":[51],"of":[52,62,70,94,115,127],"mammography":[53,97,179],"data:":[54],"(1)":[55],"its":[56],"high-res":[57,96,116],"nature,":[58],"(2)":[60],"homogeneity":[61],"radiology":[63],"reports,":[64],"largely":[65],"driven":[66],"by":[67],"a":[68,78,87,112,137,157,211],"predominance":[69],"negative/benign":[71],"findings":[72],"on":[73,184,190],"examinations.":[74],"propose":[76],"TopKSigLIP,":[77],"VLM":[79,128],"designed":[80],"address":[82,142],"these":[83],"limitations":[85],"through":[86],"novel":[88],"learning":[91],"objectives.":[92],"Instead":[93],"downscaling":[95],"images":[98],"satisfy":[100],"GPU":[101],"memory":[102],"constraints,":[103],"TopKSigLIP":[104,175,203],"introduces":[105],"TopK-Patch":[106,223],"module":[107,224],"learns":[109],"sample":[111],"sparse":[113],"set":[114],"patches":[117],"likely":[118],"contain":[120],"lesions,":[121],"sidestepping":[122],"resolution--batch":[124],"size":[125],"tradeoff":[126],"training.":[129],"The":[130,222],"sampled":[131],"patch":[132],"locations":[133],"additionally":[134,225],"serve":[135],"built-in":[138],"localization":[139,229],"tool.":[140],"To":[141],"report":[143],"homogeneity,":[144],"we":[145],"replace":[146],"contrastive":[148],"loss,":[149],"which":[150],"falsely":[151],"repels":[152],"semantically":[153],"similar":[154],"pairs,":[155],"with":[156,168],"Sup-sigmoid":[158,160],"loss.":[159],"loss":[161,165],"extends":[162],"sigmoid":[164],"from":[166,172],"SigLIP":[167],"soft":[169],"labels":[170],"derived":[171],"structured":[173],"data.":[174],"outperforms":[176],"existing":[177],"open-source":[178],"general":[181],"medical":[182],"VLMs":[183],"both":[185],"internal":[186],"external":[188],"benchmarks":[189],"density":[191],"assessment,":[192],"classification,":[194],"finding":[195],"subtyping,":[196],"cancer":[198],"prediction":[199],"under":[200,206],"evaluation.":[202],"remains":[204],"competitive":[205],"linear":[207],"probing":[208],"despite":[209],"using":[210],"significantly":[212],"smaller":[213,217],"vision":[214],"encoder":[215],"batches":[219],"than":[220],"baselines.":[221],"achieves":[226],"superior":[227],"lesion":[228],"over":[230],"post-hoc":[231],"Grad-CAM.":[232],"Code":[233],"weights":[235],"are":[236],"made":[237],"public:https://github.com/Youngseok0001/TopKSigLIP.":[238]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-05T00:00:00"}