ResMultNet-50: An automatic medical Image Diagnosis approach for lung diseases using deep transfer learning

Abstract

Lung diseases such as Covid-19, Pneumonia and Tuberculosis remains to be among the leading causes of deaths globally. These diseases present themselves in a similar manner bearing common signs and symptoms such as coughing, fever, fatigue and shortness of breaths. To prevent adverse effects of these diseases and save more lives, early detection and diagnosis of the aforementioned diseases is necessary. This paper proposes a deep transfer learning model: ResMultNet50 to assist radiologists in their work while adopting inverse class weighting approach to handle class imbalance problem. The proposed approach relied on fine-tuned ResNet-50 for the diagnostic task of detecting the three respiratory diseases from chest x-rays. In the study, a data set comprising of 13,188 chest x-ray images was used and the proposed approach achieved an average accuracy of 96.12%. This model outperformed other deep learning models and transfer learning models used in the previous studies for solving multi-class related problems.