Validation of Image Analysis Software for Breast Cancer Receptor Detection
Author Information
Author(s): Krecsák László, Micsik Tamás, Kiszler Gábor, Krenács Tibor, Szabó Dániel, Jónás Viktor, Császár Gergely, Czuni László, Gurzó Péter, Ficsor Levente, Molnár Béla
Primary Institution: Semmelweis University
Hypothesis
The study aims to validate a semi-automated image analysis software for detecting estrogen and progesterone receptors in breast cancer specimens.
Conclusion
The NuclearQuant software application is a reliable tool for pathologists to test estrogen and progesterone receptor status in breast cancer.
Supporting Evidence
- The calibrated algorithm achieved 87% agreement in object detection with manual scoring.
- Cohen's kappa showed almost perfect agreement (κ = 0.981) between the software and pathologist scores.
- Quadratic weighted kappa indicated substantial agreement (κ = 0.795) in the equivalence assessment.
- Significant correlation was found between pathologist scores and those provided by the software (Spearman's rho = 0.953).
Takeaway
This study shows that a computer program can help doctors check for important markers in breast cancer more accurately and quickly.
Methodology
The study involved calibrating the software using digital slides from breast cancer cases and comparing its scoring with that of pathologists.
Potential Biases
Potential misclassification errors may arise from poorly stained samples.
Limitations
The software may struggle with poorly prepared specimens and requires careful selection of regions of interest.
Participant Demographics
The study used breast cancer slides from 16 cases, including 14 invasive ductal carcinoma and 2 invasive lobular carcinoma.
Statistical Information
P-Value
p<0.0001
Confidence Interval
95% CI for rho 0.911-0.976
Statistical Significance
p<0.0001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website