Pathologist Computer-Aided Diagnostic Scoring of Tumor Cell Fraction: A Swiss National Study

Area of Science:

• Digital Pathology
• Computational Pathology
• Oncology

Background:

• Tumor cell fraction (TCF) estimation is crucial in cancer diagnosis but suffers from significant interobserver variability among pathologists.
• This variability can impact treatment decisions and patient outcomes, highlighting the need for objective assessment tools.

Purpose of the Study:

• To evaluate the impact of a tumor cell fraction computer-aided diagnostic (TCFCAD) tool on pathologists' TCF estimation accuracy and confidence.
• To assess whether TCFCAD support can reduce interobserver variability in TCF assessment.

Main Methods:

• Sixty-nine pathologists estimated TCF in colorectal cancer images before and after using a TCFCAD tool.
• The TCFCAD tool provided an overlay highlighting tumor vs. nontumor cells and TCF percentage.
• Pathologists reported confidence levels, and results were compared against expert-defined ground truth (GT).

Main Results:

• TCFCAD assistance significantly decreased interobserver variability (SD 9.9% vs. 5.8%, P < .0001) and improved agreement with GT.
• Intraclass correlation coefficient increased from 0.8 to 0.93, indicating better consistency.
• Pathologist confidence in their TCF estimates significantly increased when using the TCFCAD tool (3.67 to 4.17 on a 5-tier scale).

Conclusions:

• TCFCAD tools can enhance TCF estimation accuracy, interpathologist agreement, and scorer confidence in digital pathology.
• The study suggests that AI-powered tools are valuable for supporting pathologists and may increase adoption with proper training and education.