AI to Allow to Identify Patients Likely to Benefit from Immunotherapy
Immunotherapy is no doubt a novel approach in cancer treatment. However, cancer patients not only battle the disease but the steep cost that comes with it. The low success rate—just 20% may benefit—has made things worse for them as well as clinicians. Artificial intelligence (AI) for one has shown a way to measure the success rate in chemotherapy. But immunotherapy is different. Lack of understanding of the phenotype and the limited information that CT scans offer to doctors on the lesion size make tracking the therapy response difficult.
A study that involved multiple institutes sheds light on the very biology of immunotherapy—not just some one-off indications. The research at Case Western Reserve University merged medical imaging and machine learning with AI to detect patterns in CT scans to this end. The use of the system will enable oncologists to identify which patients will show a positive immunotherapy response.
The details of the study are online on the November’s issue of the journal Cancer Immunology Research.
Training AI to Track Not Just Size but also Entire Texture of Lesion
The study tracked the CT scans of patients suffering from lung cancer at different points in time. The team used AI to register the patterns in CT scans when the respondents were first diagnosed with the cancer. Then, after a few cycles of immunotherapy, the AI system again tracked the lesion.
The ML model allows oncologists, assert researchers, to not just view the size of the lesion but also the entire texture. The size can be a misguiding factor. But texture, along with volume and shape, gives a more reliable information on the therapy response. The findings confirmed that patients with most changes in their lesion had the best immune response to cancer invasion.
Patients with Most Changes in Lesion Indicated Positive Therapy Response
To make the results more reliable, the researchers used the AI tool to scan individual patients at different sites and with different immunotherapy agents. Thus they could account the variation due to different immune checkpoint inhibitors. The results confirmed that their model was indeed effective in identifying a positive response to treatment and pair it with the boost in survival rates. They now intend to test the validity of the results across different patients at different sites.
