[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-dermatology-tool-learns-not-to-crop-out-the-mole":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},8258,"ai-dermatology-tool-learns-not-to-crop-out-the-mole","AI Dermatology Tool Learns Not to Crop Out the Mole","A new routing method keeps skin-lesion detail intact when AI trims image patches to save compute, beating generic pruning on a public dataset.","A new AI training method makes sure dermatology scanners do not crop out the mole they are diagnosing.\n\nVision Transformer classifiers used to read dermoscopy images split each photo into patches and process every one, which is expensive. To cut costs, researchers often prune tokens, discarding patches judged less useful. But existing pruning methods use generic signals like saliency or patch similarity, and never check whether the discarded set still contains the lesion itself. A new paper, MedTokenBudget, adds a scoring module called Lesion-Aware Token Scoring (LATS) that combines attention entropy, feature norm, and local contrast to rank patches, then keeps only the top ones under a fixed budget. It is trained directly against ground-truth lesion masks, so it learns to protect diagnostic evidence rather than just visually busy regions.\n\nTested on the ISIC 2019 skin-lesion dataset, LATS beat both random pruning and ToMe, a widely used token-merging method, retaining substantially more actual lesion tissue at the same computational budget. That distinction matters in medical imaging, where an efficiency gain is worthless if the tradeoff is quietly cropping out the thing being diagnosed.\n\nThe catch: training this scorer requires lesion masks, hand-labeled data most real-world deployments will not have sitting around, so the efficiency win comes with an upfront labeling bill.","[\"ai\",\"computer-vision\",\"medical-ai\",\"dermatology\"]","2026-09-28T04:00:00.000Z","2026-09-28T20:05:03.314Z","2026-09-28T20:05:09.989Z","published",null,[],"ai",[24,26,27,28],"computer-vision","medical-ai","dermatology",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.30613",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,90,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",4878,"2026-09-28T16:14:54.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",764,"2026-09-28T12:00:00.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",401,"2026-09-28T15:14:48.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",268,"2026-09-28T14:00:00.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",191,"2026-09-28T15:45:00.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",138,"2026-09-28T14:13:01.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",85,"2026-09-28T13:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",78,"2026-09-28T15:29:21.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":91,"slug":92,"count":88,"latest_published_at":93},"General","general","2026-09-26T17:02:42.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",30,"2026-09-24T20:07:31.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]