AI-based tool promises improved treatment for patients with DCIS of the breast

A study published July 9 in The Lancet Digital Health reveals a new AI-based computational pathology tool designed to better manage treatment for patients with ductal carcinoma in situ (DCIS), a form of pre-invasive breast cancer. The research, led by Anant Madabhushi, PhD, cancer immunology researcher at Winship Cancer Institute of Emory University and executive director of the Emory Empathetic AI for Health Institute, promises to reduce overtreatment and improve patient outcomes with more personalized, targeted treatments by identifying which patients with DCIS are at higher risk of disease progression and would benefit most from radiation therapy.

Madabhushi emphasizes the tool's potential impact: "This AI-based assay could be a game-changer in the management of patients with DCIS, allowing clinicians to make more informed decisions and potentially sparing many patients from unnecessary radiation therapy."

Key Findings:

• Predictive Power: The AI tool, named CPath TILs, is the first of its kind and can predict which women with DCIS are likely to experience a recurrence or progression to invasive breast cancer and would benefit from radiation therapy, offering clinically actionable insights for oncologists and surgeons.
• Non-Destructive and Cost-Effective: Unlike traditional molecular assays, this AI tool is cost-effective and non-destructive to tissue samples, leveraging digital images of biopsy slides. Its lower cost makes it accessible to a broader population, including those in low- and middle-income countries and areas.
• Validated Research: The validation of this AI-based assay was conducted using data from the previously completed UK/ANZ trial, a unique 2-by-2 clinical trial designed to assess different treatment protocols and strategies for 755 women with DCIS. The main goal was to see how well the tool could predict two types of breast cancer events: recurrence of DCIS (DCIS-IBE) and progression to invasive breast cancer (I-IBE). Madabhushi and the research team performed a blinded validation of their AI classifier on biopsy slide images from the UK/ANZ trial, effectively identifying with 95% accuracy which women were at a much higher risk of progressing to invasive breast cancer and would benefit from radiation therapy. This robust validation process confirms the reliability and clinical applicability of the AI tool in making informed treatment decisions for DCIS patients.

Clinical Impact:

• This AI-based TIL score tool can help doctors better predict which patients are at higher risk of their DCIS progressing to invasive cancer and would benefit most from additional treatments like radiation therapy. This means more personalized and effective treatment plans for DCIS patients, potentially sparing some from unnecessary treatments and focusing efforts on those at higher risk.
• Patients with a high TIL score (CPath TIL-high) had more than double the risk of experiencing any breast event (recurrence or progression to invasive cancer) compared to those with a low TIL score (CPath TIL-low). The risk of progression to invasive breast cancer was more than three times higher in patients with a high TIL score. Patients with high TIL scores benefited more from radiation therapy, which significantly reduced their risk of recurrence and progression.
• For patients with low TIL scores, the AI-based tool can help avoid unnecessary radiotherapy, reducing the treatment burden and associated risks. Overtreatment could be avoided in about 15% of DCIS patients.

The study addresses a significant gap in DCIS treatment management. Traditional risk stratification methods rely on conventional parameters like tumor size and grade, which research has shown are not reliable indicators of risk, often leading to overtreatment with radiation therapy. This AI tool offers a more precise approach, combining prognostic and predictive capabilities to tailor treatment to individual patients. It has been shown to be a very strong predictor of progression to invasive disease compared to approaches previously reported in the literature, which is pivotal for preventing potentially life-threatening conditions. Prior studies of TIL biomarkers faced challenges such as bias and variability, but this research overcomes these limitations by using an AI-based approach and randomized trial data to produce robust results.

Driven by personal loss, Madabhushi has dedicated significant efforts to improving breast cancer treatment and outcomes for patients. “I lost my aunt to breast cancer over two decades ago. I have been very passionate about addressing the challenges of breast cancer for two decades. The fortuitous meeting with Dr. Thorat and Dr. Badve led to this collaboration about six years ago,” says Madabhushi.

The study is co-authored by Mangesh Thorat, MBBS, MS, FEBS, PhD, Honorary Reader, Wolfson Institute of Population Health at Queen Mary University of London, UK. Additional collaborators include Sunil Badve, MD, professor and vice-chair of the Emory Department of Pathology and Laboratory Medicine and researcher in Winship Cancer Institute’s Cell and Molecular Biology Research Program; Arpit Aggarwal, PhD student in the Biomedical Engineering program at Emory and Georgia Tech; and Haojia Li, AI scientist from Picture Health, who contributed as first author.

“We have done two key things here,” says Thorat. “Firstly, using the material from a randomized trial, we employed a very robust study design. This allowed us to eliminate limitations of previous studies and evaluate the biomarker in the best possible manner. Secondly, we harnessed the potential of AI to measure biomarkers in a very precise quantitative manner, something humans cannot easily do,” he adds. “The result is that we have a robust biomarker that not only predicts which patients are at a substantially higher risk of progressing to invasive breast cancer but also tells us which subgroup of patients can avoid radiotherapy and thus help us prevent overtreatment.”

For example, “for patients with LG/IG DCIS tumors up to 25mm, there is considerable debate about whether to use radiotherapy after surgery. These patients make up about 25% of all DCIS cases. Despite evidence showing that radiotherapy reduces recurrence but does not improve survival, many of these patients still receive radiotherapy,” Thorat explains. “This is problematic because radiotherapy can increase the risks of lung cancer, heart disease and cancer in the opposite breast.”

The team plans to explore further applications of the tool, such as aiming to spare patients from tamoxifen treatment and deploying the assay in prospective clinical trials.

This research was supported by multiple grants, including funding from the National Institutes of Health (NIH) and various philanthropic organizations dedicated to advancing cancer research.

This work is part of Emory University's broader mission to harness AI for health equity and innovation. The Emory Empathetic AI for Health Institute focuses on developing cost-effective, accessible AI tools to transform health care globally. The institute is committed to scaling these innovations to improve patient outcomes worldwide, particularly in underserved communities. The use of AI and precision medicine are projected to be an integral part of the future of personalizing and targeting cancer treatment and prevention to individuals’ specific health needs, disease progression and biological makeup.