Federated learning–enhanced particle swarm optimization for personalized low-protein diet recommendations in diabetic nephropathy patients

Abstract

Diabetic nephropathy (DN) is a major complication of diabetes, and nutritional management plays an important role in its management. Although low-protein diets (LPDs) are well documented to reduce further deterioration of renal function, patients may experience problems choosing appropriate meals for the day that can be followed with very few nutrient restrictions and without developing deficiencies. To tackle this challenge, in this study, a novel approach called federated learning (FL) combined with particle swarm optimization (PSO) is proposed. It allows the development of several nutrient prediction models at various hospitals while maintaining the privacy of patients’ data. The globally trained model is then used locally on patient’s devices, and PSO is used to create an optimized and personalized low-protein meal plan. This dual-level approach enables the system to learn from a variety of patients and to customize dietary advice to each patient. Detailed algorithmic formulations, federated averaging algorithm, and a multiobjective PSO fitness function, system flow representation is included in the study. Experimental results reveal that the proposed FL-PSO framework generates accurate, reliable, and clinically viable meal plans, which is better than the conventional centralized approaches.