Pathologist-Artificial Intelligence Concordance in HER2 Interpretation for Advanced Biliary Tract Cancer:...

Area of Science:

• Oncology
• Pathology
• Medical Imaging

Background:

• Biliary tract cancer (BTC) is an aggressive malignancy with limited treatment options.
• HER2 overexpression occurs in a subset of BTC patients, making it a target for therapy.
• Accurate HER2 assessment in BTC is challenging, particularly for low expression levels.

Purpose of the Study:

• To compare HER2 interpretations in advanced BTC by pathologists using light microscopy (LM) and digital pathology (DP).
• To evaluate the potential of artificial intelligence (AI)-powered pathology in enhancing HER2 scoring consistency.
• To quantify intraobserver and interobserver variability in HER2 evaluation for BTC.

Main Methods:

• 309 HER2 immunohistochemistry slides from advanced BTC patients were analyzed.
• Three pathologists independently evaluated HER2 expression twice using LM and DP.
• An AI whole slide image analyzer assessed HER2 expression, with ground truth determined by consensus.

Main Results:

• Pathologists achieved complete agreement in 62.1% (LM) and 63.4% (DP) of cases.
• AI showed an 83.5% concordance rate with the ground truth for HER2 categories.
• Clinical factors influenced pathologist variability, while histologic grade affected AI performance.

Conclusions:

• AI-powered HER2 scoring demonstrates high concordance with pathologist evaluations in BTC.
• AI pathology is expected to assist pathologists in achieving more objective HER2 assessments.
• Improved HER2 scoring consistency can enhance therapeutic decisions for BTC patients.