A Deep Network Model for Examining Kidney Cyst using Learning and Optimization Approaches

Due to its non-radioactive, noninvasive, real-time, and low cost, ultrasonography is frequently used to diagnose illnesses of the internal organs. Measuring markers is positioned at two distinct places to assess tumours using ultrasonography. The target finding's location and size are subsequently quantified using this information. Regardless of age, renal cysts are one of the measurement goals of abdominal ultrasonography that affect 20–50% of the population. As kidney cysts are measured from ultrasound images regularly, automating the measuring process would also have a large impact. The objective is to model a novel deep learning model that could detect kidney cysts automatically in ultrasound images and predict where two important anatomical markers should be placed to determine the cysts' sizes. The deep neural network model used optimized to predict feature maps which show the locations of salient landmarks and optimized to detect kidney cysts. Three sonographers manually marked 100 test data items that were not visible with prominent landmarks so that the results could be compared to human performance. The ground truth was a board-certified radiologist documented these key landmark points. Next, sonographers' accuracy on deep learning models are compared and analysed.