Unsupervised Feature Extraction for Various Computer-Aided Diagnosis Using Multiple Convolutional Autoencoders and 2.5-Dimensional Local Image Analysis

There are growing expectations for AI computer-aided diagnosis: diagnosis (CAD) systems can be used to improve the accuracy of diagnostic imaging. However, it is not easy collect large amounts disease medical image data with lesion area annotations supervised learning CAD systems. This study proposes an unsupervised local feature extraction method running without such datasets. Local features one key determinants system performance. The proposed requires only a normal dataset that does include lesions and collected easier than dataset. extracted by applying multiple convolutional autoencoders analyze various 2.5-dimensional images. evaluated two kinds problems: detection cerebral aneurysms in head MRA images lung nodules chest CT In both cases, performance high, showing AUC more 0.96. These results show automatically learn useful recognition from lesion-free data, regardless type or lesion.