0
0
January 16, 2025 — The advent of artificial intelligence (AI) in healthcare has brought transformative changes, particularly in the field of cancer diagnosis. A pivotal thesis by Abhinav Sharma at the Department of Medical Epidemiology and Biostatistics explores novel deep learning-based models to improve breast cancer diagnosis and treatment. Sharma’s work tackles one of the longstanding challenges in histopathology—the inter-observer and inter-lab variabilities in assessing prognostic markers, a factor that can lead to inconsistent treatments for patients.
Histopathological evaluation, which involves examining tumor specimens under a microscope, remains crucial in diagnosing breast cancer. Yet, differences in interpretation by pathologists can lead to significant clinical consequences, including under- or over-treatment. With the increasing digitization of pathology labs, computational pathology is showing promise in addressing these issues. Sharma’s research, a culmination of four comprehensive studies, presents a significant leap in AI-powered precision pathology, which could revolutionize breast cancer care.
Key Findings from Sharma’s Thesis
Sharma’s thesis introduces four innovative studies focusing on the application of deep learning models for precision pathology:
- Deep Learning for Histological Grading: Sharma’s first study developed and validated predGrade, a deep learning-based model for classifying invasive breast cancer into three grades using H&E-stained whole slide images (WSIs). The model demonstrated the potential to reduce inter-observer and inter-lab variability, providing a more reproducible and robust clinical decision-support tool for histological grading.
- Risk Stratification Using AI: In his second study, Sharma validated Stratipath Breast, an AI solution for risk stratification in breast cancer, across two independent hospital sites in Sweden. The solution significantly improved prognostic risk stratification for intermediate-risk patients, aiding in better decisions regarding adjuvant chemotherapy and reducing the risks of over- or under-treatment.
- Spatially Interpreting AI Models: Sharma’s third study introduced the Wsi rEgion sElection approach (WEEP), a methodology for interpreting the spatial decisions made by weakly supervised deep learning models. This approach provides valuable insights into how AI models arrive at conclusions, enhancing the transparency and reliability of AI-powered diagnostics.
- Multi-Stain Prognostic Prediction: The final study focused on a multi-stain deep learning model that improves prognostic risk score predictions for breast cancer patients using routinely stained WSIs. Sharma found that combining local and spatial alignment of multiple stains offered a better risk-stratification solution than individual stains alone, advancing the precision of breast cancer prognosis.
Shaping the Future of AI in Cancer Diagnostics
Sharma, who comes from an interdisciplinary background in bioengineering, emphasizes the importance of AI in healthcare. His work is a significant step toward realizing the potential of AI-based precision pathology, enabling more accurate, personalized treatment for cancer patients.
Reflecting on the future of AI in healthcare, Sharma expressed his passion for bridging biology and technology. He believes that the intersection of these fields will continue to drive innovation and hopes his research will inspire further advancements in improving diagnostics and treatment options for patients worldwide.
As AI technology progresses, the possibility of more robust, automated, and precise cancer diagnostics is on the horizon. Sharma’s work offers a glimpse into the future, where AI can play an integral role in enhancing clinical decision-making and ultimately improving patient outcomes in the battle against breast cancer.
For more information on Sharma’s research, see the full thesis: Development and validation of novel deep learning-based models for cancer histopathology image analysis (2024). DOI: 10.69622/27291567.v1.
