Robust Heatmap Template Generation for COVID-19 Biomarker Detection

Mirtha Lucas, Miguel Lerma, Jacob Furst, Daniela Raicu
2021 EAI Endorsed Transactions on Bioengineering and Bioinformatics  
INTRODUCTION: Detecting and identifying patterns in chest X-ray images of Covid-19 patients are important tasks for understanding the disease and for making differential diagnosis. OBJECTIVES: The purpose of this work is to develop a technique for detecting biomarkers of four possible conditions in chest X-rays, and study the patterns arising from the location of biomarkers. METHODS: We use transfer learning applied to a pretrained VGG19 neural network to build a model capable of detecting the
more » ... our conditions in chest X-rays. For biomarkers detection we use Grad-CAM. Patterns in the biomarkers are found by using classical eigenfaces approach. RESULTS: The discovered patterns are consistent across images from a given class of disease, and they are robust with respect to changes in dataset. CONCLUSION: The identified patterns can serve as biomarkers for a given disease in chest X-ray images, and constitute explanations of how the deep learning model makes classification decisions.
doi:10.4108/eai.24-2-2021.168729 fatcat:vp7carfbubfnxfc2deafgc6j24