Noncontrast Magnetic Resonance Radiomics and Multilayer Perceptron Network Classifier: An approach for Predicting Fibroblast Activation Protein Expression in Patients With Pancreatic Ductal Adenocarcinoma

Background Fibroblast activation protein (FAP) in pancreatic ductal adenocarcinoma (PDAC) is closely related to the prognosis and treatment of patients. Accurate preoperative FAP expression can better identify the population benefitting from FAP‐targeting drugs. Purpose To develop and validate a machine learning classifier based on noncontrast MRI for the preoperative prediction of FAP expression in patients with PDAC. Study Type Retrospective cohort study. Population Altogether, 129 patients with pathology‐confirmed PDAC undergoing MR scan and surgical resection; 90 patients in a training cohort, and 39 patients in a validation cohort. Field Strength/Sequence/3T Breath‐hold single‐shot fast‐spin echo T2‐weighted sequence and unenhanced and noncontrast T1‐weighted fat‐suppressed sequences. Assessment FAP expression was quantified using immunohistochemistry. For each patient, 1409 radiomics features were extracted from T1‐ and T2‐weighted images and reduced using the least absolute shrinkage and selection operator logistic regression algorithm. A multilayer perceptron (MLP) network classifier was developed using the training and validation set. The MLP network classifier performance was determined by its discriminative ability, calibration, and clinical utility. Statistical Tests Kaplan–Meier estimates, student's t‐test, the Kruskal–Wallis H test, and the chi‐square test, univariable regression analysis, receiver operating characteristic curve, and decision curve analysis were used. Results A log‐rank test showed that the survival of patients with low FAP expression (24.43 months) was significantly longer ( P < 0.05) than that in the FAP‐high group (13.50 months). The prediction model showed good discrimination in the training set (area under the curve [AUC], 0.84) and the validation set (AUC, 0.77). The sensitivity, specificity, accuracy, positive predictive value, and negative predictive value for the training set were 75.00%, 79.41%, 0.77, 0.86, and 0.66, respectively, whereas those for the validation set were 85.00%, 63.16%, 0.74, 0.71, and 0.80, respectively. Data Conclusions The MLP network classifier based on noncontrast MRI can accurately predict FAP expression in patients with PDAC. Evidence Level 2 Technical Efficacy Stage 2