Artificial Intelligence

Francesco Ciompi

Aims
Learn how basic methods of image processing work
Learn about important applications in the field of medical image analysis and computer aided diagnosis (CAD)
Understand the increasing role of machine learning and deep learning in medical imaging
Design, implement, and evaluate a medical image analysis system for a clinical application
Learn how to design studies to evaluate medical image analysis systems

Content
Medical imaging is increasingly gaining importance in medicine. Radiologists use images to detect diseases in an early stage (via screening), to diagnose patients with symptoms, and to monitor the effect of treatment. In pathology digitization of microscopy imaging is just starting, enabling pathologists to use computerized analysis of high-resolution gigapixel images to better diagnose disease in tissue samples. However, as the complexity of imaging (3D/4D) and the amount of data increases the interpretation of images by humans becomes problematic.
Therefore, there is a growing need for intelligent image analysis systems that can aid clinicians with image interpretation and decisions. The goal of these systems is to reproduce visual skills of highly trained human observers and to provide quantitative analysis. For this purpose, machine learning is applied to develop a computer model that can be trained to exploit information from a large number of medical images.
In recent years, Deep Learning [LeCun et al., Nature, 2015] has emerged as the state-of-the-art approach for image analysis applications. While human readers still are superior in most applications, Convolutional Neural Networks have been successfully applied to medical imaging problems like automated reading of mammograms for breast cancer detection, automatic detection of pulmonary nodules for lung cancer screening, detection of breast and prostate cancer in histopathology images and segmentation of white matter lesions in brain magnetic resonance, amongst others, de facto gradually bridging the gap between humans and computers. Finally, recent studies have shown that deep learning algorithms have reached and outperformed human professionals at diagnostic tasks like detection of skin cancer [Esteva et al., Nature, 2017], classification of diabetic retinopathy [Gulshan et al., JAMA, 2016] detection of breast cancer metastasis in lymph nodes [Ehteshami et al., JAMA, 2017], and malignancy risk estimation of lung nodules on CT [Venkadesh et al., Radiology, 2021].

In the first part of the course, students will learn basic concepts of digital image processing, medical imaging, machine learning and deep learning through a series of weekly practical assignments. Lectures and assignments will cover the following topics:
Introduction to Medical Image Analysis
Medical Image Processing and Transformation
Detection, Segmentation and Classification in Medical Imaging
Machine Learning with Neural Networks
Deep Learning with Convolutional Neural Networks
Convolutional Neural Networks and Segmentation in Medical Imaging
Convolutional Neural Networks and Detection in Medical Imaging
Deep Learning for Gand Challenges in Medical Imaging

The second part of the course is dedicated to the development of a CAD system based on deep learning, in the form of a final project, in which small teams of students will apply the concepts and techniques learned in the first part of the course and compete in ongoing grand challenges in medical imaging.