An advanced comparative analysis of proposed deep learning architectures for musculoskeletal disorder classification

Deep learning is at the cutting-edge of artificial intelligence (AI) and is emerging rapidly. Over the last few years, it has participated more significantly in medical image analysis. Hence after extensive research with many setbacks, deep learning has recently obtained significant exposures, enabling deep learning classical models to go above and beyond the human annotations of medical images. As manual assessment is often unrealistic, cumbersome and error-prone and the number of digital medical images grabbed daily is increasing rapidly. Hence there is a requirement to build a technique capable of appropriately exposing and classifying anatomy and irregularity in medical images. This paper is an advanced version of previously published work that specifically portrays the progressive performance of proposed deep learning variants such as CNN, LSTM and BiLSTM against the existing pre-trained model's results. The performance improvement has been measured by presenting an extensive comparative study for the state-of-the-art LERA dataset.