In Microsoft Health we are continuously evaluating the possibilities of machine learning methods, including image processing techniques and deep learning in medical computer vision. In this talk, we explain typical medical image analysis problems and present how we developed and evaluated deep learning methods using Python and CNTK (Cognitive Toolkit by Microsoft).
The field of medicine is underserved by technology and Microsoft Health is a research-focused incubator group leveraging AI to transform healthcare. Among areas of investigation is computer vision of medical images with Dicom format, including X-rays, CT scans, and photographs. Computer vision is an actively growing field with tremendous early impact on quality and cost of medical care. In this presentation we will show standard preprocessing techniques for computer vision of medical images using Python and CNTK (Cognitive Toolkit by Microsoft). Python is a first-class citizen of CNTK and a primary language for running deep-learning models. Through examples we demonstrate how to utilize python and libraries (skimage, OpenCV, CNTK), including image preprocessing to normalize and segment images; and deep learning model evaluation and performance. These examples will be reproducible with publicly shared code on open data sets.
https://github.com/usuyama/pydata-medical-image