ConMIL: interactive and contrastive text-guided multiple instance learning for whole slide image classification

Anxuan Han, Alexandra Jolley, Lisa M Butler, Weitong Chen, Virginie Uhlmann · 2026

Read ConMIL: interactive and contrastive text-guided multiple instance learning for whole slide image classification by Anxuan Han, Alexandra Jolley, Lisa M Butler, Weitong Chen, Virginie Uhlmann (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13614712.txt.

Description

Han and colleagues present in 2026 ConMIL, a framework that classifies whole slide images by letting written descriptions steer the computer's attention. Whole slide images are gigapixel scans of stained tissue used in pathology. Standard methods learn from slide-level labels alone and cannot say which regions drove the call.

ConMIL feeds paired positive and negative pathology text descriptions into the attention mechanism, so regions matching the described disease pattern gain weight while regions matching its opposite lose it. Human experts can refine the prompts in the loop to sharpen the wording. A special classifier network then exposes its learned curves for visual inspection. On three public slide collections covering lymph node and breast cancer, the method beat representative baselines while producing attention maps that matched pathologists' expectations of where disease sits.

Catalog record

CREATORAnxuan Han, Alexandra Jolley, Lisa M Butler, Weitong Chen, Virginie Uhlmann
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
LICENSEhttps://creativecommons.org/licenses/by/4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/article/MED/PMC13614712

Browse the full library · About the Tower of Babel