Automatic Kidney Lesion Detection for CT Images Using Morphological CNN

Abstract

The CT scan is the best tool for diagnosing and finding injuries in the kidney. It can provide precise information about the location and size of lesions in many medical applications. Manual and traditional medical tests work and time-consuming. The automatic detection of injuries in CT is now an integral task for clinical diagnosis. To develop and improve the efficiency of medical testing computer-aided diagnosis (CAD) is needed. However, the existing low accuracy and incomplete detection algorithm remain a tremendous challenge. The proposed lesion sensor is based on morphological cascaded convolutional neural networks using a multi-intersection threshold (IOU) (CNNs). To increase network stability and morphology co-detection layers and amended pyramid networks in the faster RCNN and combine four IOU threshing thresholds with cascade RCNNs and for better detection of small lesions (1-5 mm). In addition, the experiments have been conducted on CT deep-lesion kidney pictures published by photos and communication systems of hospitals (PACSs