Deep Multi-Instance Learning Using Multi-Modal Data for Diagnosis of Lymphocytosis

We investigate the use of recent advances in deep learning and propose an end-to-end trainable multi-instance convolutional neural network within a mixture-of-experts formulation that combines information from two types data-images clinical attributes-for diagnosis lymphocytosis. The learns to extract meaningful features images blood cells using embedding level approach aggregates them. Moreover, model these as well attributes form pipeline for Our results demonstrate even by itself is able discover associations between diagnosis, indicating presence important unexploited images. shown be more robust while maintaining performance via. repeatability study assess effect variability data acquisition on predictions. proposed methods are compared with different literature based both conventional handcrafted machine learning, models attention mechanisms. method reports balanced accuracy 85.41% outperfroms feature-based attention-based approaches biologists which scored 79.44%, 82.89% 77.07% respectively. These give insights potentials applicability practice. code datasets can found at https://github.com/msahasrabudhe/lymphoMIL.