The title of this book sounds very attractive because it connects two big worlds that could benefit each other. In fact, machine learning is nourished by medicine because medical problems offer new challenges that require increasingly complex methodologies of automated learning. On the other hand, medicine benefits from machine learning to support physicians making critical decisions. Thus, the reader could expect big outcomes from this book. Unfortunately, the only big thing that readers will actually get is disappointment. The two authors of the book are not computer scientists and have little background in machine learning, a term that they characterize as procedures from computing some statistical models. The result is a book that gives little or no information on the use of machine learning in medicine.
The book covers topics with a marked statistical basis, such as logistic regression, correlation analysis, linear modeling, time-dependent prediction, seasonality assessment, and factor analysis. Only a few chapters consider some typical machine learning methods such as neural networks, hierarchical clustering, or fuzzy modeling. Even with this strong bias toward statistics, the authors could have enticed interested readers to understand how these methods apply to medicine, the challenges that exist, the benefits that might accrue, and the kinds of cases that succeed. None of this appears in the book. Each chapter describes one or two methods in an extremely shallow way. The mathematical formalization is awful and the description of a model or a technique is only given by a sequence of commands for the SPSS statistical software, without any explanation of their meaning. Design problems, such as how to deal with missing data, using structured information, and so on, which are extremely common in medical situations, are simply missing from the book. The intended audience, according to the authors, includes clinicians, medical students, and clinical investigators, but I don’t believe that they could really learn how to apply machine learning to their medical problems from this text.
Finally, I found the content highly redundant. Some definitions are mechanically repeated in each chapter, and each chapter begins with a long abstract that simply replicates parts of the body. While reading the book, I was quite puzzled by the appearance of a dataset of 250 samples (with 16 features) in a six-page appendix at the end of one of the chapters (who could benefit from it?). I was astonished when I discovered that the same dataset has been repeated four times in the book.
It is very unfortunate that such a promising connection, machine learning and medicine, did not find adequate coverage in this book. I hope that the authors will do justice to this area, which has important applications and many exciting research directions, by releasing a second edition with extensive revisions.