Developing and Applying Hybrid Deep Learning Models for Computer-Aided Diagnosis of Medical Image Data

Abstract

The dissertation discusses three methods to address the challenges of applying deep learning models to medical imaging. The first method involves the development of a new joint deep learning model, J-Net, to achieve lesion segmentation and classification simultaneously. The J-Net model outperforms the individual models in accuracy with small datasets. The second method performs automatic image detection using a two-stage deep learning model to produce clean data. The third method involves developing multi-stage deep learning algorithms to generate synthetic medical image data, which can be used to overcome the lack of large, diverse datasets. These methods demonstrate that building enhanced training datasets can play a vital role in improving the performance of deep-learning models in medical imaging applications.