Application of Deep Learning in Detection of various Hepatic Disease Classification Using H & E Stained Liver Tissue Biopsy

The human liver is a vital organ that detoxifies the body and produces biochemicals that are important for body growth. Liver health is very important for a long, healthy life and the liver is prone to many ailments. Hepatic Diseases are ailments that affect the functioning of the human liver. These diseases include infections, fibrosis, ballooning or enlarging of the liver, inflammation, and others. Early diagnosis of hepatic diseases is very important and is tricky due to common symptoms. Machine Learning and Deep Learning have exhibited great potential in medical imaging and can be used to diagnose many diseases. This paper explores the application of leading deep learning models like MobileNet, Xception, and DenseNet121 that have exhibited exceptional performance on other computer vision tasks. The paper also explores the benefits of ensemble learning by building an ensemble model with the mentioned models. All models were able to perform well on a dataset of 3000 liver tissue biopsy images with 3 different classes. The average accuracy and Area Under ROC Curve scores were evaluated along with several other metrics. The average accuracy and AUROC Curve scores of MobileNet were 95.4% and 96.56% respectively while those of Xception were 95% and 96.24% respectively. DenseNet121 attained an average accuracy and AUROC Curve score of 93.74% and 95.32% while those of an ensemble model with MobileNet, DenseNet121, and Xception were 99.48% and 99.6% respectively.