U-Net-Based Assistive Identification of Bladder Cancer: A Promising Approach for Improved Diagnosis

Abstract

<b><i>Introduction:</i></b> Bladder cancer (BC) is a major health concern that poses a significant threat to the population, with an increasing incidence rate and a high risk of recurrence and progression. The primary clinical method for diagnosing BC is cystoscopy, but due to the limitations of traditional white light cystoscopy and inadequate clinical experience among junior physicians, its detection rate for bladder tumor, especially small and flat lesions, is relatively low. However, recent years have seen remarkable advancements in the application of artificial intelligence (AI) technology in the field of medicine. This has led to the development of numerous AI algorithms that have been successfully integrated into medical practices, providing valuable assistance to clinicians. The purpose of this study is to develop a cystoscopy algorithm that is real time, cost effective, high performing, and accurate, with the aim of enhancing the detection rate of bladder tumors during cystoscopy. <b><i>Materials and Methods:</i></b> For this study, a dataset of 3,500 cystoscopic images obtained from 100 patients diagnosed with BC was collected, and a deep learning model was developed utilizing the U-Net algorithm within a convolutional neural network for training purposes. <b><i>Results:</i></b> This study randomly divided 3,500 images from 100 BC patients into training and validation groups, and each patient’s pathology result was confirmed. In the validation group, the accuracy of tumor recognition by the U-Net algorithm reached 98% compared to primary urologists, with greater accuracy and faster detection speed. <b><i>Conclusion:</i></b> This study highlights the potential of U-Net-based deep learning techniques in the detection of bladder tumors. The establishment and optimization of the U-Net model is a significant breakthrough and it provides a valuable reference for future research in the field of medical image processing.