A Comparative Study of Deep Learning Models with Transfer Learning for Liver Tumor Classification

The human way of life has changed negatively over the past ten years. Living a poor lifestyle has long been a global issue. Nutrition and dietary deficiencies, stress, obesity, sleep deprivation, sleep quality, long working hours, traveling, etc., has taken a ghastly turn on most of the people’s organ health. Global statistics say that there have been 19.1 million new cancer cases in 2020 year alone. Of all the cases, liver cancers take up 8.3%. In India, according to projections by ICMR, there would be rise in cancer patients by 3 million by 2025. To study the performance of transfer learning models SqueezeNet1_1, ResNet101, and DenseNet121 for liver tumor classification is undertaken in this paper. We have used LiTS dataset for the experimentation of our proposed model. Sample LiTS Contrast Enhanced Abdominal CT Scan Images. Among the investigated three models, ResNet101 performed well with an accuracy of 95.09%. ResNet101 proved to be more sensitively promoted toward the LiTS dataset with a value of 98.34% and specificity of 97.74%. DenseNet121 proved itself with more précised classification performance with 97.12%. In future, we intend to study the performance of the ensemble models taking another step toward the challenge of liver tumor detection and classification.