After dividing image into volumes of interest (VOIs) by using multi-scale search sliding window, the VOIs are determined as tumornot through the CADe system which is trained based on deep learning architecture, ensemble learning,focal loss function. Then, the same tumor VOIs are merge by applying the hierarchical clusteringmerge algorithmthe position of merged VOI is visualized.

Different from the conventional computer-aided detection (CADe) system which is designed according to the tumor characteristic provided by experience physician, the new deep learning CADe system not only can find the appropriate learning model automatically for tumor detection, but also improve itself by more new image to achieve faster, more robust,higher accuracy purpose.

"Two modes:
1. The CADe system is applied on the scanned images directly to review the imagesdetect the suspicious lesion quickly.
2. The capture images will be saved in picture archivingcommunication system (PACS). Then, through safety networkpersonal data elimination technique, the saved images can be revieweddetected by the deployed CADe system on cloud server."