The potential of federated learning for self-configuring medical object detection in heterogeneous data distributions.

Medical Object Detection (MOD) is a clinically relevant image processing method that locates structures of interest in radiological image data at object-level using bounding boxes. High-performing MOD models necessitate large datasets accurately reflecting the feature distribution of the corresponding problem domain. However, strict privacy regulations protecting patient data often hinder data consolidation, negatively affecting the performance and generalization of MOD models. Federated Learning (FL) offers a solution by enabling model training while the data remain at its original source institution. While existing FL solutions for medical image classification and segmentation demonstrate promising performance, FL for MOD remains largely unexplored. Motivated by this lack of technical solutions, we present an open-source, self-configuring and task-agnostic federated MOD framework. It integrates the FL framework Flower with nnDetection, a state-of-the-art MOD framework and provides several FL aggregation strategies. Furthermore, we evaluate model performance by creating simulated Independent Identically Distributed (IID) and non-IID scenarios, utilizing the publicly available datasets. Additionally, a detailed analysis of the distributions and characteristics of these datasets offers insights into how they can impact performance. Our framework's implementation demonstrates the feasibility of federated self-configuring MOD in non-IID scenarios and facilitates the development of MOD models trained on large distributed databases.

The interaction of genetics and physical activity in the pathogenesis of metabolic dysfunction associated liver disease.

Genetic variants associated with increased liver fat and volume have been reported, but whether physical activity (PA) can attenuate the impact of genetic susceptibility to these traits is poorly understood. We aimed to investigate whether higher PA modify genetic impact on liver-related traits in the UK Biobank cohort. PA was self-reported, while magnetic resonance images were used to estimate liver fat (n = 27,243) and liver volume (n = 24,752). Metabolic dysfunction-associated liver disease (MASLD) and chronic liver disease (CLD) were diagnosed using ICD-9 and ICD-10 codes. Ten liver fat and eleven liver volume-associated genetic variants were selected and unweighted genetic-risk scores for liver fat (GRSLF) and liver volume (GRSLV) were computed. Linear regression analyses were performed to explore interactions between GRSLF/ GRSLV and PA in relation to liver-related traits. Association between GRSLF and liver fat was not different among lower (ß = 0.063, 95% CI 0.041-0.084) versus higher PA individuals (ß = 0.065, 95% CI 0.054-0.077, pinteraction = 0.62). The association between the GRSLV and liver volume was not different across different PA groups (pinteraction = 0.71). Similarly, PA did not modify the effect of GRSLF and GRSLV on MASLD or CLD. Our findings show that physical activity and genetic susceptibility to liver-related phenotypes seem to act independently, benefiting all individuals regardless of genetic risk.

Diet composition, nutrient substitutions and circulating fatty acids in relation to ectopic and visceral fat depots.

BACKGROUND & AIMS: Short-term randomized trials have demonstrated that replacing saturated fat (SFA) with polyunsaturated fat (PUFA) causes a reduction or prevention of liver fat accumulation, but population-based studies on diet and body fat distribution are limited. We investigated cross-sectional associations between diet, circulating fatty acids and liver fat, visceral adipose tissue (VAT), intermuscular adipose tissue (IMAT) and other fat depots using different energy-adjustment models. METHODS: Sex-stratified analyses of n = 9119 (for serum fatty acids) to 13 849 (for nutrients) participants in UK Biobank were conducted. Fat depots were assessed by MRI, circulating fatty acids by NMR spectroscopy and diet by repeated 24-h recalls. Liver fat, VAT and IMAT were primary outcomes; total adipose tissue (TAT) and abdominal subcutaneous adipose tissue (ASAT) were secondary outcomes. Three a priori defined models were constructed: the all-components model, standard model and leave-one-out model (main model including specified nutrient substitutions). Imiomics (MRI-derived) was used to confirm and visualize associations. RESULTS: In women, substituting carbohydrates and free sugars with saturated fat (SFA) was positively associated with liver fat (ß (95% CI) = 0.19 (0.02, 0.36) and ß (95% CI) = 0.20 (0.05-0.35), respectively) and IMAT (ß (95% CI) = 0.07 (0.00, 0.14) and ß (95% CI) = 0.08 (0.02, 0.13), respectively), whereas substituting animal fat with plant fat was inversely associated with IMAT, ASAT and TAT. In the all-components and standard models, SFA and animal fat were positively associated with liver fat, IMAT and VAT whereas plant fat was inversely associated with IMAT in women. Few associations were observed in men. Circulating polyunsaturated fatty acids (PUFA) were inversely associated with liver fat, IMAT and VAT in both men and women, whereas SFA and monounsaturated fatty acids were positively associated. CONCLUSIONS: Type of dietary fat may be an important determinant of ectopic fat in humans consuming their habitual diet. Plant fat and PUFA should be preferred over animal fat and SFA. This is corroborated by circulating fatty acids and overall consistent through different energy adjustment models. TWITTER SUMMARY: In UK Biobank, intake of saturated- and animal fat were positively whereas biomarkers of polyunsaturated fat were inversely associated with liver-, visceral- and intermuscular fat. Type of dietary fat may be a determinant of ectopic fat, a risk factor for cardiometabolic disease.

Erratum for: MIMIR: Deep Regression for Automated Analysis of UK Biobank MRI Scans.

[This corrects the article DOI: 10.1148/ryai.210178.].