Although machine learning has the potential to change cancer treatment completely, it is not yet being used to treat uncommon tumors like ACC. ACC has a poor prognosis and requires efficient treatments and prognostic instruments. For a study, researchers used unique AI-assisted technologies to identify metabolic patterns correlated with survival using 18F-FDG PET scans and a grading system.

Sixty-nine patients, 67% female and 33% male, median age 50 (80% metastatic [59% with lung metastases], 16% localized, and 4% NED on follow-up), with at least two 18F-FDG PET scans, were studied. For the baseline and follow-up 18F-FDG PET scans, every lesion was contoured. Using TRAQinform IQ technology (AIQ Solutions), lesions were measured, matched across timepoints, and TRAQinform Profile was computed. The following imaging characteristics were taken from each patient: basic characteristics (SUVmax, SUVmean, total lesion glycolysis, and number of lesions at baseline), basic response characteristics (baseline, follow-up, and change in basic characteristics), location characteristics (number of lesions in various organs), and heterogeneity characteristics (intrapatient heterogeneity of disease and response). Cox regression models were used to calculate the univariate predictive power of each measure’s overall survival prediction. Using three-fold cross-validation, a random survival forest was used to generate the overall survival prediction for the TRAQinform Profile. The c-index was used to assess the performance of the model.

The most effective predictor of overall survival is the overall disease burden at baseline (c-index = 0.68), followed by the disease burden at follow-up (0.65), the number of pulmonary lesions at follow-up (0.62), and the number of lesions with increasing disease burden (0.62). In addition, the responder’s vs. poor responders to the standard of care therapy may be predicted by the TRAQinform Profile (c-index = 0.76).

New techniques employing AI are being added to the arsenal for managing cancer as machine learning algorithms change quickly. They demonstrated an AI-assisted system that can forecast an ACC patient’s prognosis by analyzing the lesions found on an 18F-FDG-PET scan. It translated to a more individualized approach to the therapy alternatives.

Reference: annalsofoncology.org/article/S0923-7534(22)01877-4/fulltext