Back

Deep learning applied to pretreatment CTs provides personalized prognostic insight for lung cancer patients

|

Chicago, IL – Altis Labs, Inc. (“Altis”) is announcing the results of a study being presented by Felipe S. Torres MD, PhD of Toronto General Hospital’s Joint Department of Medical Imaging at the 2020 ASCO annual meeting. 

The study illustrates how deep learning can be applied to pretreatment computed tomography (CT) imaging to enhance 2-year survival prediction in lung cancer patients beyond traditional tumor size measurements. 

In the retrospective study, a fully automated imaging-based prognostication (IPRO) model was designed to localize the entire three-dimensional (3D) space comprising the lungs and heart, and to learn deep prognostic features using a 3D convolutional neural network. The model was trained and validated using pretreatment CTs of 1,184 stage I-IV lung cancer patients with known 2-year survival outcomes. 

The model significantly increased prognostic accuracy (C-index = 0.73, p < 0.01) compared to that of clinical TNM staging (C-index = 0.63, p < 0.01).

More accurate prognostic insight derived from standard-of-care imaging not only has the potential to facilitate optimal treatment decisions but also offers cancer researchers more sophisticated tools to refine clinical trial design and quantify treatment effect.

View the full publication here.

About Altis Labs

Altis Labs is the computational imaging company advancing precision medicine. We believe that medical imaging is the richest source of untapped information. Life sciences companies use our software platform, Nota, to accelerate all stages of clinical development. Trained on over 140 million real-world images with associated diagnostics, treatment information, and outcomes, Nota predicts clinically meaningful outcomes from baseline and follow-up scans to more accurately stratify patients and quantify treatment effect. Altis is proudly based in Toronto, Canada.

To learn more, visit www.altislabs.com | info@altislabs.com