AI displays breast cancer neoadjuvant chemotherapy response features out of multi-stain histopathologic images

February 07, 2023 | Histopathology

The prediction of cancer patient outcomes and overall survival rate have been realized due to computational pathology algorithms and tools. Basing from features or grades obtained from tumor histopathologic images, it can help us recognize the potential hazard factors while improving treatment planning towards precision oncology.

However, there are still challenges brought by poor understanding of tumor immune micro-environments. Zhi Huang and colleagues performed a study where they described IMage-based Pathological REgistration and Segmentation Statistics (IMPRESS), an extensive, reproducible whole slide image feature extraction pipeline. The researchers looked into whether AI-based algorithms from automatic feature extraction methods could detect neoadjuvant chemotherapy outcomes in HER2-positive (HER2+) and triple-negative breast cancer (TNBC) patients by using both H&E and multiplex IHC images. In training the machine learning models in accurately predicting the response to NAC in breast cancer patients, features from tumor immune micro-environment and clinical data.

The results exhibited that the method outperformed the results that were manually generated by pathologists. Externally validated by independent cohorts, the developed image features and algorithms produced encouraging results, particularly for the HER2+ subtype. This implied that compared to classic clinical scores, AI can objectively assess slides and distinguish between high and low risk of relapse for early ER + HER2- breast cancer, aiding in ruling out a decent amount of cases.

This study enabled the application of computational pathology for clinical diagnosis and prognosis and the interpretation of the roles of different cellular components in tumor immune micro-environment, all of which have been discovered to play major roles in clinical outcomes of cancer.

Nevertheless, the capacity of AI may be limited to only several cancer subtypes. Thus the results of this study on HER2+ and TNBC subtypes can inspire future research outside power analysis and model selection.