From Pixels to Prognosis Machine Learning Solutions for Critical Healthcare Challenges

Abstract

Application of machine learning methods to two important problems, namely, detection of COVID- newline19 using chest radiographs (X-rays and CT scans), and molecular subtyping of breast cancer using newlinemulti-omics data is carried out. The recent pandemic made clear the need for fast and reliable techniques newlinein distinguishing pneumonia caused by the novel virus SARS-CoV-2 from pneumonia caused newlineby other viral/bacterial/fungal infections. In this work, a basic CNN model was built from scratch and newlinespatial attention-based mechanism (Attn-CNN) incorporated to detect the manifestations of COVID-19 newlinein CXR and CT scan images with improved generalizability and explainability has been developed. The newlineproposed spatial attention-based solution overcomes the need for lung segmentation and region-based newlineannotations for training the CNN models while keeping the model complexity minimized, thus making newlineit deployable in clinical settings. To verify the generalizability of the models, testing has also been newlinecarried out on external datasets and explainability has been provided using Grad-CAM visualization of newlinethe pixels, selected by the model for classification. Performance evaluation of the proposed approach newlineagainst five state-of-the-art deep learning models showed 95% accuracy for CXRs and 96% for CT newlineimages and outperformed all other models and comparatively generalized well on external datasets. newlineAdvancements in the high-throughput techniques have generated large volumes of data, enabling newlinegenome-wide profiling of various omics data, such as protein-coding and non-coding (e.g., miRNA, newlinelncRNA, etc.) genes, DNA methylation, and analysis of genetic variations (SNVs, CNVs, etc.). However, newlineidentification of diagnostic and prognostic biomarkers is challenging due to heterogeneity at multiple newlinelevels and the huge number of features associated with each. This heterogeneity is seen to affect newlinethe generalizability and explainability of ML models. To address the high dimensionality and explainability newlineissues, a knowledge-based feature selection framework along