دانلود کتاب یادگیری ماشینی و عمیق در انکولوژی فیزیک پزشکی و رادیولوژی

This book is based on one of the most consequential emergent results of the ongoing computer revolution, namely, that computers can be trained—under the right conditions— to reliably classify new data, such as patient data. This capability, called machine learning (or statistical learning), has been deployed in many areas of technology, commerce, and medicine. Data mining and statistical prediction models have already crept into many areas of modern life, including advertising, banking, sports, weather prediction, politics, science generally, and medicine in particular. The ability of computers to increasingly communicate with people in a natural way (understanding language and speaking to us), such as the famed IBM “Watson” appearance on Jeopardy, or “Siri” on iPhones, portends an accelerating role of sophisticated computer models that predict and respond to our requests. Fundamentally, these developments rely on the ability of statistical computer methods to pull (as Nate Silver puts it) “signal from the noise.” While traditional statistical methods typically attempt to ascertain the role of particular variables in determining an outcome of interest (hence, needing many data points for every variable included in the prediction model), machine learning represents a different goal, to reliably predict an outcome, for example that an imaging abnormality is benign with a high degree of certainty. The statisticians and computer scientists working in this emerging area are often happy to use large numbers of variables (or previous data instances) that essentially vote together in a nonlinear fashion. Simplicity is happily traded for an improved ability to predict.