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

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Urological diagnostics

Background:

• Bladder cancer (BC) presents a significant global health challenge with increasing incidence and recurrence rates.
• Current cystoscopy methods have limitations in detecting small or flat tumors, particularly with less experienced clinicians.
• Advancements in AI offer potential solutions for enhancing medical diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop a real-time, cost-effective, and accurate AI algorithm for cystoscopy.
• To improve the detection rate of bladder tumors during cystoscopy procedures.
• To assist clinicians in identifying bladder cancer lesions more effectively.

Main Methods:

• A dataset of 3,500 cystoscopic images from 100 BC patients was curated.
• A deep learning model was developed using the U-Net algorithm within a convolutional neural network.
• The model was trained and validated on the collected dataset.

Main Results:

• The U-Net algorithm achieved 98% accuracy in tumor recognition within the validation group.
• The AI model demonstrated superior accuracy and faster detection speeds compared to primary urologists.
• Pathology results confirmed the accuracy of the model's tumor identification.

Conclusions:

• U-Net-based deep learning shows significant potential for enhancing bladder tumor detection via cystoscopy.
• The developed U-Net model represents a breakthrough in medical image processing for urological applications.
• This research provides a valuable reference for future advancements in AI-driven medical diagnostics.