{"version":1,"type":"story","url":"https://digestai.news/story/foundation-models-match-but-sometimes-lag-task-specific-models-in-pros","json":"https://digestai.news/story/foundation-models-match-but-sometimes-lag-task-specific-models-in-pros.json","markdown":"https://digestai.news/story/foundation-models-match-but-sometimes-lag-task-specific-models-in-pros.md","slug":"foundation-models-match-but-sometimes-lag-task-specific-models-in-pros","headline":"Foundation models match but sometimes lag task-specific models in prostate pathology","summary":"The study evaluated more than 100,000 prostate biopsy images from 6,268 patients across 15 sites, comparing self‑supervised foundation models (FMs) with task‑specific (TS) end‑to‑end models. Both model families reached high agreement with pathologists, achieving quadratic Cohen’s kappa of 0.91 (95 % CI 0.90‑0.92) on internal data. On external cohorts, TS models scored 0.87 (0.86‑0.88) while the UNI FM and Virchow2 FM recorded 0.86 (0.84‑0.87) and 0.85 (0.84‑0.86) respectively.\n\nWhen sufficient labeled data were available, TS models converged with or surpassed FMs, reducing misgrading, misdiagnosis of challenging morphologies, and technical variability. However, FMs consumed up to 35 times more energy than TS models, raising concerns about resource efficiency. The authors suggest that both approaches be considered for robust, clinically applicable AI pathology solutions.\n\nThe work was supported by multiple European research grants and acknowledges shareholders in Clinsight AB and A3P Biomedical AB among the authors. The article is licensed under a Creative Commons Attribution 4.0 International License.","keyPoints":["Using over 100,000 prostate biopsies from 6,268 patients across 15 sites, the study compared foundation models and task‑specific models.","Task‑specific models achieved Cohen’s kappa up to 0.91 internally and 0.87 externally, matching or exceeding foundation models.","Foundation models required up to 35× more energy than task‑specific models while offering no clear accuracy advantage."],"whyItMatters":"Demonstrates that large foundation models may not outperform specialized models in medical imaging, guiding resource allocation and sustainability decisions in AI pathology.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":["Clinsight AB","A3P Biomedical AB","Innovation Fund Denmark","Nordforsk","Business Finland","Swedish Research Council"],"models":["UNI FM","Virchow2 FM"],"people":["N. Mulliqi","A. Blilie","X. Ji","K. K."]},"firstPublishedAt":"2026-10-26T00:00:00Z","updatedAt":"2026-10-26T00:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"Nature Machine Learning","title":"Comparative analysis of end-to-end learning and foundation models for clinical-grade prostate pathology","url":"https://nature.com/articles/s41467-026-78996-7","publishedAt":"2026-10-26T00:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Foundation models match but sometimes lag task-specific models in prostate pathology\", 26 October 2026, https://digestai.news/story/foundation-models-match-but-sometimes-lag-task-specific-models-in-pros","publisher":"Digest AI","title":"Foundation models match but sometimes lag task-specific models in prostate pathology","datePublished":"2026-10-26T00:00:00Z","url":"https://digestai.news/story/foundation-models-match-but-sometimes-lag-task-specific-models-in-pros"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}