Performance of a HER2 testing algorithm tailored for urothelial bladder cancer: A Bi-centre study

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• HER2 (Human Epidermal growth factor Receptor 2) is a crucial biomarker in several cancers.
• Accurate HER2 scoring in urothelial bladder cancer (UBCa) is essential for treatment decisions.
• Current manual HER2 scoring methods can be subjective and vary between observers.

Purpose of the Study:

• To develop and validate an AI algorithm for automated HER2 scoring in UBCa.
• To evaluate the impact of AI assistance on interobserver agreement among pathologists.
• To assess the applicability of breast cancer HER2 scoring criteria to UBCa.

Main Methods:

• Development of an AI algorithm using 330 UBCa slides from two institutions.
• A ring study involving six pathologists (3 senior, 3 junior) using 200 selected slides.
• Comparison of manual scoring versus AI-assisted scoring for accuracy and interobserver agreement (kappa statistic).

Main Results:

• The AI algorithm demonstrated high accuracy (0.94 and 0.92) in independent tests.
• AI-assisted scoring significantly improved accuracy (0.66 vs 0.94) and consistency (kappa=0.48 vs 0.87) compared to manual scoring.
• Interpretation accuracy improved substantially in HER2-low (F1-scores: 0.63 vs 0.92) and heterogeneous HER2-positive cases (0.49 vs 0.93) with AI assistance.

Conclusions:

• This is the first AI quantification algorithm for HER2 scoring in UBCa, aiding pathologist diagnosis.
• HER2 scoring criteria established for breast cancer are effectively applicable to UBCa.
• AI assistance enhances diagnostic accuracy and interobserver consistency for pathologists of all experience levels, particularly in complex cases.