AI extracts imaging data to predict lung cancer treatment response

Experts have harnessed the vast amounts of information contained in computed tomography scans to develop artificial intelligence that’s capable of predicting how patients will respond to lung cancer therapy.

A team of clinicians from Columbia University Irving Medical Center in New York included CT data from numerous phase 2 and phase 3 clinical trials to produce their machine learning model. And when tested in a validation dataset, it accurately predicted tumor-based treatment sensitivity.

Using these data-rich radiomic signatures could drastically change clinical decision-making for patients suffering from non-small cell lung cancer, the authors explained March 20 in Clinical Cancer Research.

“With AI, cancer imaging can move from an inherently subjective tool to a quantitative and objective asset for precision medicine approaches," Laurent Dercle, MD, PhD, an associate research scientist in Columbia’s Department of Radiology, said in a statement.

Currently for radiologists to assess how patients are responding to systemic therapy, they must calculate changes in tumor size and spot the appearance of new lesions. This, the authors noted, is often limited, especially when dealing with newer treatments such as immunotherapy.

Operating under the knowledge that such modern day treatments require more up-to-date approaches, Dercle et al. turned to data. Specifically, to non-small cell lung cancer individuals treated with either an immunotherapeutic agent nivolumab, chemotherapeutic agent docetaxel, or targeted therapeutic gefitinib.

Radiomic features were extracted from patients’ CT scans taken at baseline and during their initial treatment. In total, each individual had more than 1,000 features extracted from one of their lung lesions. Ultimately, the researchers settled on using eight tumor characteristics for their prediction model, including changes in tumor volume, heterogeneity, shape, and margin.

And when tested, the predictions proved to be accurate. Using an area under the curve scale in which a 1 is perfect, the nivolumab, docetaxel, and gefitinib prediction models achieved a score of 0.77, 0.67, and 0.82 in the validation dataset, respectively.

“We observed that similar radiomics features predicted three different drug responses in patients with NSCLC," Dercle said. "Further, we found that the same four features that identified EGFR treatment sensitivity for patients with metastatic colorectal cancer could be utilized to predict treatment sensitivity for patients with metastatic NSCLC."