Accuracy of breast density assessment using artificial intelligence by convolutional neural network for carriers of UGT1A1 polymorphisms with Gilbert's Syndrome - a pilot study.

introductionGilbert's Syndrome (GS), an indirect hyperbilirubinemia resulting from the reduced hepatic bilirubin-glucuronyltransferase enzyme activity, influences estrogen metabolism. Polymorphisms in the UGT gene, such as those associated with GS, can alter estrogen clearance, impacting hormonal exposure and, consequently, breast density, an independent risk factor for breast cancer. Furthermore, the accumulation of estrogenic metabolites like 4-OH-estrogens, recognized as carcinogenic agents, suggests a potential increased risk of breast cancer in GS patients. Given this complex interrelationship, this pilot study aimed to evaluate the accuracy of a Convolutional Neural Network (CNN) in determining breast density in postmenopausal women with GS, seeking to optimize their follow-up.

methodsThe authors conducted a prospective, cross-sectional study involving 21 patients, divided into GS and genotyped wild-type control groups. The CNN's performance was compared against radiologists' assessments.

resultsFor dense breasts in the GS group, the CNN showed 55.6% sensitivity, 100.0% specificity, and 77.8% accuracy. Kappa agreement was 0.381 for the GS group and 0.700 for controls (fair and good agreement, respectively). Specificity was the most representative diagnostic parameter for dense breasts.

conclusionThis pilot study provides preliminary evidence suggesting that the CNN's sensitivity in breast density assessment in women with GS may be low, while specificity seems to be high. These findings suggest the need for further validation of artificial intelligence algorithms in specific populations, such as those with GS, and that conclusions should be interpreted with caution due to the small sample size and wide confidence intervals.