AI-based physique composition evaluation predicts lung most cancers therapy outcomes

Tafadzwa Chaunzwa, MD, a researcher within the Synthetic Intelligence in Drugs (AIM) Program at Mass Normal Brigham and a senior resident doctor on the Harvard Radiation Oncology Program, is the lead creator of a paper revealed in JAMA Oncology. Chaunzwa and senior creator Hugo Aerts, PhD, director of the AIM Program, and affiliate professor at Harvard College, shared highlights from their paper.

How would you summarize your examine for a lay viewers?

As remedies like immunotherapy enhance most cancers survival charges, there’s a rising want for medical decision-support instruments that predict therapy response and affected person outcomes. That is notably essential for lung most cancers, which stays the highest reason behind most cancers dying globally. Earlier research linked physique mass index (BMI) with lung most cancers outcomes and immunotherapy drug negative effects. Nevertheless, BMI is a restricted measure that does not seize particulars about completely different physique tissues and their interplay with most cancers therapies. We used medical imaging and synthetic intelligence (AI) to research physique composition in a big cohort of sufferers with lung most cancers that has unfold to different elements of the physique. Our examine discovered that modifications in muscle mass and fats high quality throughout therapy are essential indicators of outcomes for this inhabitants.

What information gaps does your examine assist to fill?

As we proceed to enhance the therapy of superior lung most cancers with completely different systemic brokers, together with immunotherapy medicine, biomarkers which might be each prognostic and predictive of therapy response are more and more wanted to tell medical selections. Prior research recognized an affiliation between BMI and lung most cancers outcomes. An affiliation between BMI and the incidence of negative effects with immunotherapy has additionally been elucidated. Nevertheless, BMI alone is a crude metric that doesn’t seize the distribution and relative contributions of various physique tissues. Medical imaging-based analyses of physique composition are being more and more explored; nevertheless, within the setting of superior non-small cell lung most cancers (NSCLC), research have been restricted by small pattern sizes and guide and difficult-to-reproduce methodologies.

How did you conduct your examine?

We got down to carry out complete physique composition evaluation in giant cohorts of people handled for superior or metastatic lung most cancers with completely different systemic medicine. We developed a strong end-to-end AI-based platform to help with this activity.

What did you discover?

We discovered that whereas the distribution of various tissue compartments at the beginning of cancer-directed therapy had some worth, the change in these measurements over the course of therapy was extra strongly related to affected person outcomes. Particularly, we discovered that loss in muscle mass was a poor prognostic think about sufferers handled with chemotherapy, immunotherapy, or chemoimmunotherapy. Curiously, amongst sufferers receiving immunotherapy, both alone or together with chemotherapy, modifications within the high quality of the fats beneath the pores and skin (subcutaneous adipose tissue), as seen on CT scans, had been additionally related to the danger for lung most cancers development or mortality.

What are the implications?

This examine presents key breakthroughs that may assist advance the prognostication and surveillance of sufferers receiving immunotherapy for NSCLC. The primary breakthrough is the implementation of an automatic AI-based pipeline for complete physique composition evaluation at scale in a various inhabitants of sufferers receiving immunotherapy and cytotoxic chemotherapy for superior NSCLC. That is the most important and most intensive such examine, incorporating each real-world information and potential medical trial cohorts, with longitudinal assortment of multimodal information and prolonged follow-up to observe illness response to remedy. Our outcomes reveal the potential of this evaluation framework to supply a extra nuanced understanding of the connection between physique composition and response to immunotherapy in NSCLC in comparison with crude BMI measurements. This will have essential medical implications for affected person choice, therapy, and monitoring. The second contribution is sharing this sturdy end-to-end deep-learning pipeline for automated slice choice and physique compartment segmentation on cross-sectional imaging.

What are the following steps?

We provide the software program as an open-source AI instrument that seamlessly integrates with platforms for picture evaluation and in addition disseminate it on the modelhub.ai platform. By making this algorithm publicly accessible, we hope to speed up future research on this area and additional facilitate the event of novel approaches for analyzing complicated most cancers imaging information units. This may permit additional investigation of essential associations established utilizing BMI or guide CT-based physique composition measurements. Extra broadly, advances on this analysis space will assist information the administration of various cancers and enhance our capability for precision oncology.