Stefano Ghidoni – författare

This book introduces readers to the current trends in using deep learners and deep learner descriptors for medical applications. It reviews the recent literature and presents a variety of medical image and sound applications to illustrate the five major ways deep learners can be utilized: 1) by training a deep learner from scratch (chapters provide tips for handling imbalances and other problems with the medical data); 2) by implementing transfer learning from a pre-trained deep learner and extracting deep features for different CNN layers that can be fed into simpler classifiers, such as the support vector machine; 3) by fine-tuning one or more pre-trained deep learners on an unrelated dataset so that they are able to identify novel medical datasets; 4) by fusing different deep learner architectures; and 5) by combining the above methods to generate a variety of more elaborate ensembles. This book is a value resource for anyone involved in engineering deep learners for medical applications as well as to those interested in learning more about the current techniques in this exciting field. A number of chapters provide source code that can be used to investigate topics further or to kick-start new projects.