With foundation-model features, aggregator complexity does not improve cross-hospital discrimination in gastric whole-slide image classification
Organizations: College of AI Convergence, Institute of Digital Anti-Aging Healthcare, u-AHRC, Inje University, 197 Inje-ro, Gimhae, 50834, Gyeongsangnam-do, Republic of Korea. · Department of Pathology, Dankook University Hospital, 201 Manghyang-ro, Dongnam-gu, Cheonan, 31116, Chungcheongnam-do, Republic of Korea. · Department of Pathology, Ewha Womans University Seoul Hospital, 260 Gonghang-daero, Gangseo-gu, Seoul, 07804, Republic of Korea. · Department of Pathology, Keimyung University Dongsan Medical Center, 1035 Dalgubeol-daero, Dalseo-gu, Daegu, 42601, Republic of Korea. ·
Abstract
Whole-slide image classifiers are usually developed within one hospital and often go untested at unseen hospitals. We assembled 5,036 haematoxylin-and-eosin-stained gastric slides from five Korean hospitals, scanned at 20x and 40x, and compared multiple-instance learning aggregators under leave-one-hospital-out validation on frozen pathology foundation-model features. A hierarchical patch-region-slide graph aggregator (CHIME-MIL) with site alignment and two-scale fusion reached a mean area under the receiver operating characteristic curve (AUC) of 0.837, above six conventional attention and transformer aggregators trained without these steps. Under a matched recipe it did not exceed attention-based pooling, and global max pooling given the same alignment and fusion was better (AUC 0.855, P = 0.0002). The alignment, which subtracts each hospital's mean feature vector without labels, onboarded a new hospital from ten unlabelled slides and cost 0.002 AUC without held-out-hospital data, but its accuracy gains were small and inconsistent. The model had low top-label calibration error (0.053) but under-confident malignancy probabilities, and removing its most-attended patches lowered the malignancy probability far more than removing random patches. With these features, more complex aggregators did not improve discrimination; what a deployable system adds is label-free onboarding, calibration checks and auditable evidence.
Histopathology Image Classification
Histopathology image classification uses computational models to identify tissue types, tumors, grades, and biomarkers from stained pathology images.
See all 151 papers