Core Concepts
A deep learning-based approach utilizing pyramid sampling for the automated classification of HER2 status in immunohistochemically-stained breast cancer tissue images.
Abstract
The study introduces a deep learning-based method that utilizes pyramid sampling to automate the classification of HER2 status in immunohistochemically (IHC)-stained breast cancer tissue images. The pyramid sampling strategy captures morphological features at various spatial scales, efficiently managing the computational load and facilitating a detailed examination of cellular and larger-scale tissue-level details. This method addresses the tissue heterogeneity of HER2 expression by providing a comprehensive view, leading to a blind testing classification accuracy of 84.70%, on a dataset of 523 core images from tissue microarrays. The automated system has the potential to enhance diagnostic precision and evaluation speed, and might significantly impact cancer treatment planning.
The study first provides an overview of the challenges in manual HER2 evaluation, including inter- and intra-observer inconsistency and extended turnaround times. It then describes the proposed deep learning-based approach, which utilizes a pyramid sampling strategy to capture multi-scale tissue features. The classification network is trained on a dataset of 1462 core images from 823 patients, with an additional 162 cores from 149 patients used for validation. The model's performance is then blindly evaluated using 523 core images from 300 patients.
The results demonstrate the capability of the automated HER2 scoring system through qualitative and quantitative analyses. The model's predictions reflect the level of HER2 expression, with the majority of high-confidence predictions aligning with the consensus category in most samples. Monte Carlo simulations are used to optimize the number of independent pyramid sampling sets (N) and the confidence threshold parameter (k) for the final HER2 score prediction. The study shows that increasing the number of PSSs (N) up to 20 significantly improves the classification accuracy, which reaches a maximum of 87.76%. Additionally, the analysis of the confidence threshold parameter (k) reveals that a value within the range of 1 to 20 maintains optimal classification performance.
The discussion highlights the key advantages of the proposed approach, including its ability to address the challenge of HER2 expression heterogeneity and its potential to standardize HER2 assessment, streamline pathologists' workflow, and improve diagnostic accuracy. The study concludes that the automated, deep learning-based HER2 scoring framework represents a significant advancement in breast cancer diagnostics, contributing to the advancement of personalized medicine and improving patient care in oncology.
Stats
Blind testing set includes 523 core images from 300 patients that were not previously seen by the model during the training or validation.
The training set consists of 1462 core images from 823 patients, and the validation set includes 162 cores from 149 patients.
Quotes
"Our automated system, proving reliable as an adjunct pathology tool, has the potential to enhance diagnostic precision and evaluation speed, and might significantly impact cancer treatment planning."
"By prioritizing predictions with the highest confidence, our approach reduces the chance of inaccuracies due to lower-confidence inferences, ensuring the final HER2 score is based on significant expressions."
"This automated approach can help standardize HER2 assessment, streamline pathologists' workflow, and improve diagnostic accuracy."