Abstract
Abstract
Automated triage of whole-slide cervical cytology images remains constrained by the heavy dependence on single-cell bounding boxes or pixel-level annotations, which introduce significant bottlenecks, high expert overhead, and observer variability. To address these challenges, we present a weakly-supervised, detection-free Multiple Instance Learning (MIL) framework powered by a Gated Attention pooling engine that operates exclusively on slide-level predictions. By treating whole-slide images as bags of local instance patches, the dual-branch gated network dynamically assigns non-linear attention weights to highlight isolated dysplastic cells while suppressing benign background and debris, successfully resolving the “needle-in-a-haystack” problem inherent to high-grade lesions. Evaluated across multi-center cohorts including internal datasets (SIPaKMeD, Herlev, and CRIC) and the unannotated, out-of-distribution Mendeley LBC validation cohort processed via an unsupervised marker-controlled watershed pipeline our approach demonstrates robust generalization without requiring dense instance labels. Furthermore, our architectural comparison reveals a key trade-off: lightweight backbones like MobileNetV2 optimize in-distribution multi-center accuracy (90.96% Accuracy, 0.9800 ROC-AUC), whereas higher-capacity models like Xception ensure superior out-of-distribution robustness (80.43% Balanced Accuracy) under domain shifts. Ultimately, this detection-less pipeline provides a scalable and clinically viable solution to automate slide triage and alleviate cytopathologist shortages in resource-constrained environments.
Author summary
Automated screening of cervical cytology slides is traditionally hindered by the need for labor-intensive, pixel-level cell annotations. To overcome this limitation, this study introduces a weakly-supervised, detection-free Multiple Instance Learning framework that performs slide triage using only slide-level diagnostic labels. By employing a Gated Attention mechanism, the model dynamically focuses on informative cellular regions while ignoring background debris, successfully addressing the challenge of identifying rare dysplastic cells across gigapixel images. Evaluations across multi-center datasets demonstrate that lightweight models like MobileNetV2 achieve high accuracy on in-distribution data, whereas deeper architectures like Xception provide superior robustness against out-of-distribution variations. Ultimately, this approach offers a scalable, annotation-free triage tool designed to assist cytopathologists and improve screening workflows in resource-constrained environments.