Site Editor
Gregory J. Riely, MD, PhD
Gregory J. Riely, MD, PhD
Posted: Thursday, January 12, 2023
Artificial intelligence (also known as deep learning) may be capable of predicting genetic abnormalities in non–small cell lung cancer (NSCLC), according to research presented in JCO Clinical Cancer Informatics. In a feasibility study, Akira Ono, MD, of Shizuoka Cancer Center, Japan, and colleagues used a commercially available deep learning platform and whole-slide imaging data to predict the presence of ALK gene rearrangement in NSCLC.
“If a model with greater accuracy can be achieved, it will expand the possibilities for companion diagnostics for druggable gene abnormalities by using artificial intelligence and reduce the cost, labor, and time required to transport histopathologic specimens,” noted the authors.
The study included 208 cases of NSCLC, 66 with and 142 without ALK rearrangement. Cases were diagnosed between January 2009 and March 2019 via immunohistochemical staining. A total of 300 digital slides, 150 with and 150 without ALK rearrangement, were created with NanoZoomer and analyzed using HALO artificial intelligence. A learning model was generated using the DenseNet network. Slides were randomly assigned to either the training (n = 172) or testing (n = 128) cohort. The training cohort built the prediction models, and the testing cohort evaluated prediction performance. Data from the testing cohort were also evaluated by an expert pathologist to independently determine whether ALK rearrangement was present.
Four resolutions—16.0, 4.0, 1.0, and 0.25 μm/pix—were tested. In the 1.0 μm/pix resolution, the maximum area under the curve was 0.73. With a rearrangement probability of a 50% threshold, sensitivity and specificity were each 73%. In the same cohort, the expert pathologist detected a sensitivity of 13% and a specificity of 94%.
Disclosure: For full disclosures of the study authors, visit ascopubs.org.
JCO Clinical Cancer Informatics
JNCCN Spotlights
