Kamel, Ihab R.

ABSTRACTBackgroundDeep learning (DL) models for accurate renal tumor characterization may benefit from multi‐center datasets for improved generalizability; however, data‐sharing constraints necessitate privacy‐preserving solutions like federated learning (FL).PurposeTo assess the performance and reliability of FL for renal tumor segmentation and classification in multi‐institutional MRI datasets.Study TypeRetrospective multi‐center study.PopulationA total of 987 patients (403 female) from six hospitals were included for analysis. 73% (723/987) had malignant renal tumors, primarily clear cell carcinoma (n = 509). Patients were split into training (n = 785), validation (n = 104), and test (n = 99) sets, stratified across three simulated institutions.Field Strength/SequenceMRI was performed at 1.5 T and 3 T using T2‐weighted imaging (T2WI) and contrast‐enhanced T1‐weighted imaging (CE‐T1WI) sequences.AssessmentFL and non‐FL approaches used nnU‐Net for tumor segmentation and ResNet for its classification. FL‐trained models across three simulated institutional clients with central weight aggregation, while the non‐FL approach used centralized training on the full dataset.Statistical TestsSegmentation was evaluated using Dice coefficients, and classification between malignant and benign lesions was assessed using accuracy, sensitivity, specificity, and area under the curves (AUCs). FL and non‐FL performance was compared using the Wilcoxon test for segmentation Dice and Delong's test for AUC (p < 0.05).ResultsNo significant difference was observed between FL and non‐FL models in segmentation (Dice: 0.43 vs. 0.45, p = 0.202) or classification (AUC: 0.69 vs. 0.64, p = 0.959) on the test set. For classification, no significant difference was observed between the models in accuracy (p = 0.912), sensitivity (p = 0.862), or specificity (p = 0.847) on the test set.Data ConclusionFL demonstrated comparable performance to non‐FL approaches in renal tumor segmentation and classification, supporting its potential as a privacy‐preserving alternative for multi‐institutional DL models.Evidence Level4.Technical EfficacyStage 2.

Background Screening individuals at increased risk for pancreatic cancer (PC) detects early, potentially curable, pancreatic neoplasia. Objective To develop consortium statements on screening, surveillance and management of high-risk individuals with an inherited predisposition to PC. Methods A 49-expert multidisciplinary international consortium met to discuss pancreatic screening and vote on statements. Consensus was considered reached if ≥75% agreed or disagreed. Results There was excellent agreement that, to be successful, a screening programme should detect and treat T1N0M0 margin-negative PC and high-grade dysplastic precursor lesions (pancreatic intraepithelial neoplasia and intraductal papillary mucinous neoplasm). It was agreed that the following were candidates for screening: first-degree relatives (FDRs) of patients with PC from a familial PC kindred with at least two affected FDRs; patients with Peutz–Jeghers syndrome; and p16 , BRCA 2 and hereditary non-polyposis colorectal cancer (HNPCC) mutation carriers with ≥1 affected FDR. Consensus was not reached for the age to initiate screening or stop surveillance. It was agreed that initial screening should include endoscopic ultrasonography (EUS) and/or MRI/magnetic resonance cholangiopancreatography not CT or endoscopic retrograde cholangiopancreatography. There was no consensus on the need for EUS fine-needle aspiration to evaluate cysts. There was disagreement on optimal screening modalities and intervals for follow-up imaging. When surgery is recommended it should be performed at a high-volume centre. There was great disagreement as to which screening abnormalities were of sufficient concern to for surgery to be recommended. Conclusions Screening is recommended for high-risk individuals, but more evidence is needed, particularly for how to manage patients with detected lesions. Screening and subsequent management should take place at high-volume centres with multidisciplinary teams, preferably within research protocols.

Objective Pancreatic cysts are commonly detected in patients undergoing pancreatic imaging. Better approaches are needed to characterise these lesions. In this study we evaluated the utility of detecting mutant DNA in secretin-stimulated pancreatic juice. Design Secretin-stimulated pancreatic juice was collected from the duodenum of 291 subjects enrolled in Cancer of the Pancreas Screening trials at five US academic medical centres. The study population included subjects with a familial predisposition to pancreatic cancer who underwent pancreatic screening, and disease controls with normal pancreata, chronic pancreatitis, sporadic intraductal papillary mucinous neoplasm (IPMN) or other neoplasms. Somatic GNAS mutations (reported prevalence ∼66% of IPMNs) were measured using digital high-resolution melt-curve analysis and pyrosequencing. Results GNAS mutations were detected in secretin-stimulated pancreatic juice samples of 50 of 78 familial and sporadic cases of IPMN(s) (64.1%), 15 of 33 (45.5%) with only diminutive cysts (<5 mm), but none of 57 disease controls. GNAS mutations were also detected in five of 123 screened subjects without a pancreatic cyst. Among 97 subjects who had serial pancreatic evaluations, GNAS mutations detected in baseline juice samples predicted subsequent emergence or increasing size of pancreatic cysts. Conclusion Duodenal collections of secretin-stimulated pancreatic juice from patients with IPMNs have a similar prevalence of mutant GNAS to primary IPMNs, indicating that these samples are an excellent source of mutant DNA from the pancreas. The detection of GNAS mutations before an IPMN is visible suggests that analysis of pancreatic juice has the potential to help in the risk stratification and surveillance of patients undergoing pancreatic screening.