Predicting response to neoadjuvant therapy using artificial intelligence on digitized histopathology slides: a...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Neoadjuvant therapy (NAT) is crucial in surgical oncology, but patient response varies significantly.
• Predicting NAT response is vital for optimizing treatment strategies and patient outcomes.

Purpose of the Study:

• To systematically review and evaluate artificial intelligence (AI) models for predicting NAT response.
• To analyze AI methodologies, data modalities, and NAT types used in predicting treatment outcomes from H&E-stained biopsy slides.

Main Methods:

• Systematic literature search across five databases adhering to PRISMA guidelines.
• Quality assessment of included studies using QUADAS-2 criteria.
• Analysis of 25 eligible studies focusing on AI model performance and validation.

Main Results:

• AI models demonstrated promising predictive performance, with reported Area Under the Curve (AUC) values typically between 0.70 and 0.90.
• Approximately 40% of the reviewed studies incorporated external validation cohorts, enhancing model generalizability.
• Heterogeneity in AI methodologies and data acquisition was noted across studies.

Conclusions:

• AI models hold significant potential for predicting neoadjuvant therapy response from histopathological images.
• Future research should prioritize standardized data practices, patient-level validation, and data/code sharing for robust clinical implementation.
• Enhancing data transparency and reproducibility is essential for advancing AI in oncologic pathology.