|
Abstract:
|
There has been an explosive growth in the interest as well as applications of medical imaging technologies in all phases of cancer management, including the diagnosis, staging, treatment planning, postoperative surveillance, and response evaluation, in the last few decades. Now, quantitative imaging is an essential component of cancer clinical protocols and is able to furnish morphological, structural, metabolic and functional information associated with cancer tumors. Deep learning algorithms, in particular convolutional networks, have rapidly become a state-of-the-art choice for all tasks of quantitative imaging analysis, including image classification, object detection, segmentation, registration, and other tasks. We provide an understanding of deep learning from a metricization building perspective. This allows to rethink some of key issues associated with various deep learning methods, such as architecture design in imaging analysis. We end with a critical discussion of open challenges and directions for future research. This is based on a series of works with students and postdoctoral fellows in BIGS2 lab.
|
