This book looks at one of the most important challenges mankind has faced since it was first identified thousands of years ago: the fight against cancer. There is no question that cancer is one of the major causes of mortality worldwide. And there is no question that one of the most important technologies in the human era is artificial intelligence (AI). This book combines these two aspects, that is, how AI-based techniques can be applied to cure cancer; in other words, how AI can contribute to better computer-aided diagnostics (CAD).
The book opens with a comprehensive analysis of why cancer is one of the most significant causes of human mortality. This first chapter analyzes the different kinds of cancer from an epidemiology standpoint; even more importantly, it analyzes how those types of cancer are detected with “classical” techniques. The introductory chapter closes with a list of well-known citations that allow interested readers to feed their curiosity.
The central part of the book (chapters 2 to 5) shows examples of well-known deep learning techniques for detecting different kinds of cancers: oral cancer (chapter 2), breast cancer (chapter 3), lung cancer (chapter 4), and skin cancer (chapter 5). It is important to remark that all the chapters used open-source and freely available datasets (Kaggle) as well as open-source software libraries (scikit-learn, Keras, TensorFlow) for the programming language Python.
Chapter 2 details three deep convolutional neural network (CNN) pipelines based on VGG16 to identify oral squamous cell carcinoma (OSCC). Results show higher accuracy for the pipeline combining VGG16 with a random forest (93 percent) versus the pipeline using a support vector machine and k-nearest neighbor (kNN). The experiments used 990 OSCC histopathological images.
Chapter 3 focuses on two feature-based transfer learning pipelines for classifying breast cancer. It first describes the (public and free) dataset used for running the experiments (that is, the BreaKHis 400X dataset) and then analyzes the importance of the tuning parameters of the two compared approaches (for example, kernel functions).
Chapter 4 deals with lung cancer. In this case, the performance of a deep learning neural network is tested against four categories of lung cancer. As in the previous chapter, several pipelines and configurations are evaluated, resulting in the DenseNet121-SVM pipeline with an accuracy of 87 percent. As in the previous case, the dataset used for analysis comes from an open-source repository (Kaggle), which will allow interested readers to replicate the results.
Chapter 5 is related to skin cancer diagnostics. This chapter attempts to classify three classes of skin cancers (two malignant and one benign). To accomplish this, the authors employ two VGG-logistic regression pipelines and another pipeline combining the previous ones. The study shows that the combined pipeline yields the best accuracy (78 percent versus 74 percent and 73 percent in the testing dataset).
Chapter 6 summarizes the main contributions of the previous chapters and gives solid references for future investigations.
In this short but precise book, each chapter includes a broad state-of-the-art section and compares the performances of several AI CAD approaches to the most common cancers using freely available datasets. In addition to the previous key aspects, those approaches are implemented using open-source software tools (Python, scikit-learn, Keras, and TensorFlow).
This book is intended for AI professionals and medical teams who are responsible for CAD approaches in healthcare settings, as well as researchers and PhD students in the areas of computer science (CS) engineering and medicine.
