Reliable detection of lymph nodes in whole pelvic for radiotherapy

Accurate detection of the lymph nodes in patients with cervical cancer enables more targeted treatment. For drainage or radiotherapy of postoperative adjuvant, clinicians will manually label the lymph nodes in Computed tomography (CT) images. However, due to the complex environment and wide area of pelvic, reliable detection of lymph nodes is challenging because of the changeable shapes, sizes, distributions. Furthermore, the confusion of blood vessels and tissues in CT images also causes difficulties when labeling lymph nodes. In this study, a reliable convolutional neural network (CNN) based detection approach is proposed to distinguish lymph nodes in the middle abdomen and pelvic cavity between CT image sequences. Combining both local and global contextual information in CT image sequences can avoiding complex three-dimensional (3D) computation. Different scale transformation and structural similarity-based processing method ensure the accuracy to distinguish lymph nodes and blood vessels. Experimental results compared with those given by radiologists and other CNN based methods proved that our proposed method could locate lymph nodes with 98.29% accuracy and 94.64% recall rate when detecting 22,846 clinical abdominal CT images, which proved its potential application value for radiotherapy of cervical cancer.