Obtaining spatially resolved tumor purity maps using deep multiple instance learning in a pan-cancer study.

Tumor purity is the percentage of cancer cells within a tissue section. Pathologists estimate tumor purity to select samples for genomic analysis by manually reading hematoxylin-eosin (H&E)-stained slides, which is tedious, time consuming, and prone to inter-observer variability. Besides, pathologists' estimates do not correlate well with genomic tumor purity values, which are inferred from genomic data and accepted as accurate for downstream analysis. We developed a deep multiple instance learning model predicting tumor purity from H&E-stained digital histopathology slides. Our model successfully predicted tumor purity in eight The Cancer Genome Atlas (TCGA) cohorts and a local Singapore cohort. The predictions were highly consistent with genomic tumor purity values. Thus, our model can be utilized to select samples for genomic analysis, which will help reduce pathologists' workload and decrease inter-observer variability. Furthermore, our model provided tumor purity maps showing the spatial variation within sections. They can help better understand the tumor microenvironment.

Artificial intelligence modelling in differentiating core biopsies of fibroadenoma from phyllodes tumor.

Breast fibroepithelial lesions (FEL) are biphasic tumors which consist of benign fibroadenomas (FAs) and the rarer phyllodes tumors (PTs). FAs and PTs have overlapping features, but have different clinical management, which makes correct core biopsy diagnosis important. This study used whole-slide images (WSIs) of 187 FA and 100 PT core biopsies, to investigate the potential role of artificial intelligence (AI) in FEL diagnosis. A total of 9228 FA patches and 6443 PT patches was generated from WSIs of the training subset, with each patch being 224 × 224 pixel in size. Our model employed a two-stage architecture comprising a convolutional neural network (CNN) component for feature extraction from the patches, and a recurrent neural network (RNN) component for whole-slide classification using activation values from the global average pooling layer in the CNN model. It achieved an overall slide-level accuracy of 87.5%, with accuracies of 80% and 95% for FA and PT slides respectively. This affirms the potential role of AI in diagnostic discrimination between FA and PT on core biopsies which may be further refined for use in routine practice.