Deep learning-enhanced MRI for diabetic foot tarsal bone lesions and insulin injection behavior analysis

Diabetic foot (DF) complications, including tarsal bone lesions, are a major cause of morbidity in patients with diabetes, often leading to severe disability. Accurate diagnosis through advanced imaging techniques and understanding patient self-care behaviors, such as insulin injection, are critical for effective management. This study aimed to explore tarsal bone lesions in DF using magnetic resonance imaging (MRI) enhanced by an improved denoising convolutional neural network (DnCNN) denoising algorithm and to investigate factors influencing self-insulin injection behavior in DF patients. A total of 89 patients with diabetes were divided into a control group (42 cases without complicated DF) and a DF group (47 cases with complicated DF). MRI scans were conducted, and images were processed using the improved DnCNN algorithm. Cuboid bone lesions were compared between groups, and a customized questionnaire was used to analyze self-injection behavior. The peak signal-to-noise ratio (PSNR) did not differ significantly between the original DnCNN and the improved DnCNN ( P > 0.05), but the improved DnCNN had a significantly shorter processing time ( P < 0.05). The DF group had a higher incidence of single bone marrow edema, bone marrow edema with fracture, plantar ulcer, sinus tract, soft tissue swelling, and joint dislocation ( P < 0.05). The injection technique score in the DF group was lower than in other dimensions ( P < 0.05), with nonstandard rates for alcohol disinfection, discarded needle disposal, and site rotation exceeding 75 %. A positive correlation was found between self-injection scores and psychological insulin resistance ( r = 0.43, P < 0.05). The improved DnCNN algorithm enhances MRI diagnostic accuracy for DF-related bone and soft tissue changes, while the high rate of irregularities in self-injection techniques highlights the need for targeted educational interventions to improve patient outcomes.