Fully convolutional network and sparsity-based dictionary learning for liver lesion detection in CT examinations

In this work we focus on liver metastases detection in computed tomography (CT) examinations, using both a global context with a fully convolutional network (FCN), and a local patch level analysis with superpixel sparse based classification. The task of detecting metastases in the liver, in particular the small metastases, is important for early detection of liver cancer. Using a combined global and local approach, we present a system that can enhance detection capabilities. Our data contains a development set with CT examinations from 20 patients with a total of 68 lesions and a testing set with CT examinations from 14 patients with overall 55 lesions, out of which 35% were considered small lesions (longest diameter ≤ 1.5 cm). Experiments using 3-fold cross-validation resulted in a true positive rate of 94.6% with 2.9 false positives per case. These results are clinically promising, and should lead to better detection capabilities, including of small lesions, which is critical in cancer diagnosis.