@This paper introduces and reviews machine learning techniques in computer-aided diagnosis (CAD) schemes for three kinds of CAD: detection of lung nodules, and diagnosis of same in computed tomography (CT) images, and detection of polyps in CT colonography (CTC). There are some survey papers on CAD systems for specific organs. This paper makes a unique contribution in that it focuses on comparisons of machine learning techniques in computer-aided detection and diagnosis schemes in thoracic CT and CTC. The author divides machine learning techniques used in CAD into three classes. Feature-based machine learning techniques such as SVM and ensemble learning are the most commonly used techniques. However, the author also makes the point that pixel-based machine learning techniques, which directly use pixel/voxel/patch values in images as features, are practical. The authorfs proposal is an example of a pixel-based machine learning massive training artificial neural network (MTANN). MTANNs and their variants are applicable to all three kinds of CAD and are an effective way to improve the performance of CAD schemes.
@This paper is very useful and valuable for both CAD researchers and the machine learning community. |
