METHOD: Two large datasets, including 1110 3D CT images, were split into five segments of 20% each. Each dataset's first 20% segment was separated as a holdout test set. 3D-CNN training was performed with the remaining 80% from each dataset. Two small external datasets were also used to independently evaluate the trained models.
RESULTS: The total combination of 80% of each dataset has an accuracy of 91% on Iranmehr and 83% on Moscow holdout test datasets. Results indicated that 80% of the primary datasets are adequate for fully training a model. The additional fine-tuning using 40% of a secondary dataset helps the model generalize to a third, unseen dataset. The highest accuracy achieved through transfer learning was 85% on LDCT dataset and 83% on Iranmehr holdout test sets when retrained on 80% of Iranmehr dataset.
CONCLUSION: While the total combination of both datasets produced the best results, different combinations and transfer learning still produced generalizable results. Adopting the proposed methodology may help to obtain satisfactory results in the case of limited external datasets.
METHOD: Medical image data for five types of defects were selected, segmented, converted and decimated to 3D polygon models on a personal computer. The models were transferred to a computer aided design (CAD) software which aided in designing the prosthesis according to the virtual models. Two templates were designed for each defect, one by an OS (free) system and one by CS. The parameters for analyses were the virtual volume, Dice similarity coefficient (DSC) and Hausdorff's distance (HD) and were executed by the OS point cloud comparison tool.
RESULT: There was no significant difference (p > 0.05) between CS and OS when comparing the volume of the template outputs. While HD was within 0.05-4.33 mm, evaluation of the percentage similarity and spatial overlap following the DSC showed an average similarity of 67.7% between the two groups. The highest similarity was with orbito-facial prostheses (88.5%) and the lowest with facial plate prosthetics (28.7%).
CONCLUSION: Although CS and OS pipelines are capable of producing templates which are aesthetically and volumetrically similar, there are slight comparative discrepancies in the landmark position and spatial overlap. This is dependent on the software, associated commands and experienced decision-making. CAD-based templates can be planned on current personal computers following appropriate decimation.
METHOD: For designing and modeling the DSPN severity grading systems for MNSI, 19 years of data from Epidemiology of Diabetes Interventions and Complications (EDIC) clinical trials were used. Different Machine learning-based feature ranking techniques were investigated to identify the important MNSI features associated with DSPN diagnosis. A multivariable logistic regression-based nomogram was generated and validated for DSPN severity grading using the best performing top-ranked MNSI features.
RESULTS: Top-10 ranked features from MNSI features: Appearance of Feet (R), Ankle Reflexes (R), Vibration perception (L), Vibration perception (R), Appearance of Feet (L), 10-gm filament (L), Ankle Reflexes (L), 10-gm filament (R), Bed Cover Touch, and Ulceration (R) were identified as important features for identifying DSPN by Multi-Tree Extreme Gradient Boost model. The nomogram-based prediction model exhibited an accuracy of 97.95% and 98.84% for the EDIC test set and an independent test set, respectively. A DSPN severity score technique was generated for MNSI from the DSPN severity prediction model. DSPN patients were stratified into four severity levels: absent, mild, moderate, and severe using the cut-off values of 17.6, 19.1, 20.5 for the DSPN probability less than 50%, 75%-90%, and above 90%, respectively.
CONCLUSIONS: The findings of this work provide a machine learning-based MNSI severity grading system which has the potential to be used as a secondary decision support system by health professionals in clinical applications and large clinical trials to identify high-risk DSPN patients.
METHODS: The model was developed to estimate knee torque from experimentally derived MMG signals and other parameters related to torque production, including the knee angle and stimulation intensity, during NMES-assisted knee extension.
RESULTS: When the relationship between the actual and predicted torques was quantified using the coefficient of determination (R2), with a Gaussian support vector kernel, the R2 value indicated an estimation accuracy of 95% for the training subset and 94% for the testing subset while the polynomial support vector kernel indicated an accuracy of 92% for the training subset and 91% for the testing subset. For the Gaussian kernel, the root mean square error of the model was 6.28 for the training set and 8.19 for testing set, while the polynomial kernels for the training and testing sets were 7.99 and 9.82, respectively.
CONCLUSIONS: These results showed good predictive accuracy for SVR modelling, which can be generalized, and suggested that the MMG signals from paralyzed knee extensors are a suitable proxy for the NMES-assisted torque produced during repeated bouts of isometric knee extension tasks. This finding has potential implications for using MMG signals as torque sensors in NMES closed-loop systems and provides valuable information for implementing this method in research and clinical settings.