Search papers, labs, and topics across Lattice.
This paper introduces an attention-based multiple instance learning (ABMIL) framework for predicting the predominant lung adenocarcinoma (LUAD) growth pattern from whole slide images (WSIs), reducing the need for extensive pixel-level annotations. The method leverages pretrained pathology foundation models (frozen or fine-tuned) to encode patches, which are then aggregated using attention mechanisms to predict the predominant growth pattern. Results demonstrate that fine-tuning the patch encoders, particularly Prov-GigaPath, within the ABMIL framework achieves a substantial agreement (\k{appa} = 0.699) with expert annotations.
Skip the pixel-perfect annotations: attention-based MIL with pathology foundation models can predict lung cancer growth patterns from whole slide images with surprisingly high accuracy.
Lung adenocarcinoma (LUAD) grading depends on accurately identifying growth patterns, which are indicators of prognosis and can influence treatment decisions. Common deep learning approaches to determine the predominant pattern rely on patch-level classification or segmentation, requiring extensive annotations. This study proposes an attention-based multiple instance learning (ABMIL) framework to predict the predominant LUAD growth pattern at the whole slide level to reduce annotation burden. Our approach integrates pretrained pathology foundation models as patch encoders, used either frozen or fine-tuned on annotated patches, to extract discriminative features that are aggregated through attention mechanisms. Experiments show that fine-tuned encoders improve performance, with Prov-GigaPath achieving the highest agreement (\k{appa} = 0.699) under ABMIL. Compared to simple patch-aggregation baselines, ABMIL yields more robust predictions by leveraging slide-level supervision and spatial attention. Future work will extend this framework to estimate the full distribution of growth patterns and validate performance on external cohorts.
