LiteMIL: a computationally efficient cross-attention multiple instance learning for cancer subtyping on whole-slide images

PurposeAccurate cancer subtyping is essential for precision medicine but challenged by the computational demands of gigapixel whole-slide images (WSIs). Although transformer-based multiple instance learning (MIL) methods achieve strong performance, their quadratic complexity limits clinical deployment. We introduce LiteMIL, a computationally efficient cross-attention MIL, optimized for WSIs classification.ApproachLiteMIL employs a single learnable query with multi-head cross-attention for bag-level aggregation from extracted features. We evaluated LiteMIL against five baselines (mean/max pooling, ABMIL, MAD-MIL, and TransMIL) on four TCGA datasets (breast: n=875, kidney: n=906, lung: n=958, and TUPAC16: n=821) using nested cross-validation with patient-level splitting. Systematic ablation studies evaluated multi-query variants, attention heads, dropout rates, and architectural components.ResultsLiteMIL achieved competitive accuracy (average 83.5%), matching TransMIL, while offering substantial efficiency gains: 4.8× fewer parameters (560K versus 2.67M), 2.9× faster inference (1.6s versus 4.6s per fold), and 6.7× lower Graphics Processing Unit (GPU) memory usage (1.15 GB versus 7.77 GB). LiteMIL excelled on lung (86.3% versus 85.0%), TUPAC16 (72% versus 71.4%), and matched kidney performance (89.9% versus 89.7%). Ablation studies revealed task-dependent multi-query performance benefits: Q=4 versus Q=1 improved morphologically heterogeneous tasks (breast/lung +1.3% each, p