# Artificial intelligence diagnostics for bladder tumor identification and grade prediction depend on narrow band imaging cystoscopy

## TL;DR

This paper introduces an AI system that helps identify bladder cancer and predict tumor grade using narrow band imaging cystoscopy, improving diagnostic accuracy and reducing workload.

## Contribution

The AINCDS system introduces a dual-channel feature extraction and feature-sharing strategy for efficient bladder cancer diagnosis and grade prediction.

## Key findings

- AINCDS achieved 0.919 accuracy in identifying bladder cancer.
- The system predicted tumor grade with 0.764 accuracy.
- Urologists with 1–3 years’ experience improved grade prediction accuracy from 0.667 to 0.793 with AINCDS.

## Abstract

The effective treatment of bladder cancer depends on early evaluation through cystoscopy. Given the clinical importance of distinguishing the tumor grade, we report the application of the AI-assisted NBI Cystoscopy Diagnostic System (AINCDS). The AINCDS consists of (1) dual-channel feature extraction module, (2) lesion segmentation module based on feature pyramids, and (3) a multi-task classification module. AINCDS achieved an accuracy for identifying bladder cancer of 0.919 (95% CI = 0.896 to 0.938). For the prediction of tumor grade, the accuracy was 0.764 (95% CI = 0.714 to 0.810). The AINCDS demonstrates similar ability comparable to urologists with over 10 years’ experience. With the assistance of AINCDS, the tumor grade prediction accuracy of urologists with 1–3 years’ experience improved from 0.667 to 0.793. AINCDS can assist in the diagnosis of bladder cancer and prediction of tumor grade, offering the potential to improve the accuracy of lesion assessment and reduce the workload of urologists.

•AINCDS identifies bladder cancer and predicts tumor grade with NBI•AINCDS proposes a dual-channel feature extraction strategy•AINCDS incorporates a feature-sharing strategy to achieve computational efficiency

AINCDS identifies bladder cancer and predicts tumor grade with NBI

AINCDS proposes a dual-channel feature extraction strategy

AINCDS incorporates a feature-sharing strategy to achieve computational efficiency

Oncology; Medical imaging; Artificial intelligence applications