Objective:To investigate the feasibility of thigh muscles segmentation in patients with Duchenne muscular dystrophy (DMD) using the TransUNet deep learning model on MRI.Methods:This was a cross-sectional study. From April 2023 to September 2024, the axial T 1WI imagings of 60 DMD patients, confirmed by genetic analysis at Northwest Women′s and Children′s Hospital, were enrolled in this retrospective study. Patients were divided into a training set ( n=48) and a test set ( n=12) at a ratio of 8∶2 using random sampling. Fat infiltration scores were assigned to 11 thigh muscles, including the rectus femoris, vastus lateralis, vastus intermedius, vastus medialis, sartorius, adductor longus, adductor magnus, gracilis, semimembranosus, semitendinosus, and biceps femoris long head. A total of 1 078 images were included (884 for training, 194 for testing).The 12 DMD patients in the test set were divided into groups G1 to G4, with 2, 4, 2, and 4 cases respectively, according to the total score of muscle fat infiltration from low to high. A TransUNet model was trained on the T 1WI to perform automatic segmentation of the 11 thigh muscles in both thighs. The segmentation performance of the TransUNet automatic segmentation model was evaluated using the Dice similarity coefficient (DSC), intersection over union (IoU), and average symmetric surface distance (ASSD), with the results of manual delineation by physicians as the gold standard. And one-way analysis of variance or the Kruskal-Wallis H test was used to compare the segmentation effects of the automatic segmentation model on the thigh muscles of children among G1 to G4. Results:The mean processing time for the automatic segmentation of all 11 muscles in both thighs per patient was (8.3±1.5) s. The DSC, IoU and ASSD in training set and in test set were 0.971±0.011, 0.948±0.019, 0.69 (0.55, 0.96) and 0.944±0.021, 0.900±0.038, 0.58 (0.55, 0.91), respectively. In the training set, the semitendinosus muscle achieved the best segmentation results (DSC 0.99, IoU 0.97, ASSD 0.52). In the test set, the sartorius muscle showed the best segmentation performance (DSC 0.96, IoU 0.93, ASSD 0.50). There were no statistically significant differences in the overall DSC, IoU, or ASSD of the automatic segmentation model across groups G1 to G4 ( P>0.05). Conclusions:The TransUNet automatic segmentation model can rapidly and accurately segment the bilateral thigh muscles in patients with DMD, and the segmentation performance demonstrated consistent among patients with different degrees of muscle fat infiltration.

Objective:To investigate the feasibility of thigh muscles segmentation in patients with Duchenne muscular dystrophy (DMD) using the TransUNet deep learning model on MRI.Methods:This was a cross-sectional study. From April 2023 to September 2024, the axial T 1WI imagings of 60 DMD patients, confirmed by genetic analysis at Northwest Women′s and Children′s Hospital, were enrolled in this retrospective study. Patients were divided into a training set ( n=48) and a test set ( n=12) at a ratio of 8∶2 using random sampling. Fat infiltration scores were assigned to 11 thigh muscles, including the rectus femoris, vastus lateralis, vastus intermedius, vastus medialis, sartorius, adductor longus, adductor magnus, gracilis, semimembranosus, semitendinosus, and biceps femoris long head. A total of 1 078 images were included (884 for training, 194 for testing).The 12 DMD patients in the test set were divided into groups G1 to G4, with 2, 4, 2, and 4 cases respectively, according to the total score of muscle fat infiltration from low to high. A TransUNet model was trained on the T 1WI to perform automatic segmentation of the 11 thigh muscles in both thighs. The segmentation performance of the TransUNet automatic segmentation model was evaluated using the Dice similarity coefficient (DSC), intersection over union (IoU), and average symmetric surface distance (ASSD), with the results of manual delineation by physicians as the gold standard. And one-way analysis of variance or the Kruskal-Wallis H test was used to compare the segmentation effects of the automatic segmentation model on the thigh muscles of children among G1 to G4. Results:The mean processing time for the automatic segmentation of all 11 muscles in both thighs per patient was (8.3±1.5) s. The DSC, IoU and ASSD in training set and in test set were 0.971±0.011, 0.948±0.019, 0.69 (0.55, 0.96) and 0.944±0.021, 0.900±0.038, 0.58 (0.55, 0.91), respectively. In the training set, the semitendinosus muscle achieved the best segmentation results (DSC 0.99, IoU 0.97, ASSD 0.52). In the test set, the sartorius muscle showed the best segmentation performance (DSC 0.96, IoU 0.93, ASSD 0.50). There were no statistically significant differences in the overall DSC, IoU, or ASSD of the automatic segmentation model across groups G1 to G4 ( P>0.05). Conclusions:The TransUNet automatic segmentation model can rapidly and accurately segment the bilateral thigh muscles in patients with DMD, and the segmentation performance demonstrated consistent among patients with different degrees of muscle fat infiltration.

Objective To establish a mini-column centrifugation-HPLC method to determine the entrapment efficiency of levodopa-loaded PEGylated-solid lipid nanoparticles.Methods A dextran gel(Sephadex G-50) mini-column centrifugation was employed to separate the free drug from solid lipid nanoparticles.The content of levodopa was qualified by HPLC.Results Under the applied chromatographic condition,the excipients had no influence on the determination of levodopa.A calibrated linear of levodopa concentration was within 10.54-527.00 μg·mL-1.The recoveries of high,medium and low concentrations of levodopa were 99.13%,99.51% and 99.04%(RSD were 1.25%,1.91% and 1.71%), respectively.The free levodopa was well separated from solid lipid nanoparticles by using mini-column centrifugation.The addition of blank solid lipid nanoparticles recovery was 98.84% with RSD of 0.80%(n=3).The average adsorption rates of the three concentrations of free levodopa were 100.00%,98.75% and 98.56%(RSD were 0.00%,0.19% and 0.18%,n=3),respectively.The adsorption rate of the physical mixtures of three different concentrations of drugs and empty PEGylated solid lipid nanoparticles were 99.68%,98.46% and 99.21%(RSD were 1.52%,0.23% and 0.21%),respectively.Conclusion The method was simple,accurate and reproducible,which can be used for determination of the entrapment efficiency of levodopa-loaded PEGylated-solid lipid nanoparticles.