Site Editor

Gregory J. Riely, MD, PhD


Using Deep Learning–Based Framework to Classify Lung Cancer Histology and Prognosis

By: Celeste L. Dixon
Posted: Friday, March 1, 2024

Researchers continue to ascertain how a novel deep learning–based framework called tissue-metabolomic-radiomic–computed tomography (TMR-CT), which is derived from deep probabilistic canonical correlation analysis, is potentially clinically relevant: It is being studied in the use of classifying the histology of non–small lung cancer (NSCLC) tissue and assessing a patient’s prognosis using CT alone. Eric O. Aboagye, PhD, PharmD, of Imperial College London, and colleagues described their work in npj Precision Oncology.

Employing an international CT data set of 742 patients with NSCLC, the team was able to show that histology classification and prognosis results of CT used with TMR-CT were superior to those of conventional radiomics in patients who had the CT scan alone available. TMR-CT was able to noninvasively determine histologic class (adenocarcinoma vs squamous cell carcinoma) with an F1 score of 0.78 and to assert patients’ prognosis with a C-index of 0.72.

Currently, explained Dr. Aboagye and co-investigators, “the rich chemical information from tissue metabolomics provides a powerful means to [distinguish] tissue physiology or tumor characteristics at cellular and tumor microenvironment levels. However, the process of obtaining such information requires invasive biopsies, is costly, and can delay clinical patient management.” On the other hand, “CT is a clinical standard of care but does not intuitively harbor histologic or prognostic information.”

The authors described TMR-CT as combining 48 paired CT images and tumor/normal tissue metabolite intensities to generate 10 image embeddings, with which metabolite-derived representation from CT alone can be inferred. This use of the ability to embed metabolome information into CT “surpass[es] the performance of radiomics models and deep learning on single-modality CT feature extraction,” noted the researchers. “Additionally, our work shows the potential to generate informative, biology-inspired, CT-led features to explore connections between hard-to-obtain tissue metabolic profiles and routine lesion-derived image data.”

Disclosure: The study authors reported no conflicts of interest.

By continuing to browse this site you permit us and our partners to place identification cookies on your browser and agree to our use of cookies to identify you for marketing. Read our Privacy Policy to learn more.