Automatic Detection of Meniscus Tears Using Backbone Convolutional Neural Networks on Knee MRI.

BACKGROUND: Timely diagnosis of meniscus injuries is key for preventing knee joint dysfunction and improving patient outcomes because it decreases morbidity and facilitates treatment planning. PURPOSE: To train and evaluate a deep learning model for automated detection of meniscus tears on knee magnetic resonance imaging (MRI). STUDY TYPE: Bicentric retrospective study. SUBJECTS: In total, 584 knee MRI studies, divided among training (n = 234), testing (n = 200), and external validation (n = 150) data sets, were used in this study. The public data set MRNet was used as a second external validation data set to evaluate the performance of the model. SEQUENCE: A 3 T, coronal, and sagittal images from T1-weighted proton density (PD) fast spin-echo (FSE) with fat saturation and T2-weighted FSE with fat saturation sequences. ASSESSMENT: The detection system for meniscus tear was based on the improved YOLOv4 model with Darknet-53 as the backbone. The performance of the model was also compared with that of three radiologists of varying levels of experience. The determination of the presence of a meniscus tear from surgery reports was used as the ground truth for the images. STATISTICAL TESTS: Sensitivity, specificity, prevalence, positive predictive value, negative predictive value, accuracy, and receiver operating characteristic curve were used to evaluate the performance of the detection model. Two-way analysis of variance, Wilcoxon signed-rank test, and Tukey's multiple tests were used to evaluate differences in performance between the model and radiologists. RESULTS: The overall accuracies for detecting meniscus tears using our model on the internal testing, internal validation, and external validation data sets were 95.4%, 95.8%, and 78.8%, respectively. One radiologist had significantly lower performance than our model in detecting meniscal tears (accuracy: 0.9025 ± 0.093 vs. 0.9580 ± 0.025). DATA CONCLUSION: The proposed model had high sensitivity, specificity, and accuracy for detecting meniscus tears on knee MRIs. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY: Stage 2.

Construction and Application of Rib Fracture Diagnosis Model Based on YOLOv3 Algorithm.

OBJECTIVES: The artificial intelligence-aided diagnosis model of rib fractures based on YOLOv3 algorithm was established and applied to practical case to explore the application advantages in rib fracture cases in forensic medicine. METHODS: DICOM format CT images of 884 cases with rib fractures caused by thoracic trauma were collected, and 801 of them were used as training and validation sets. A rib fracture diagnosis model based on YOLOv3 algorithm and Darknet53 as the backbone network was built. After the model was established, 83 cases were taken as the test set, and the precision rate, recall rate, F1-score and radiology interpretation time were calculated. The model was used to diagnose a practical case and compared with manual diagnosis. RESULTS: The established model was used to test 83 cases, the fracture precision rate of this model was 90.5%, the recall rate was 75.4%, F1-score was 0.82, the radiology interpretation time was 4.4 images per second and the identification time of each patient's data was 21 s, much faster than manual diagnosis. The recognition results of the model was consistent with that of the manual diagnosis. CONCLUSIONS: The rib fracture diagnosis model in practical case based on YOLOv3 algorithm can quickly and accurately identify fractures, and the model is easy to operate. It can be used as an auxiliary diagnostic technique in forensic clinical identification.

Recognition Method of Knob Gear in Substation Based on YOLOv4 and Darknet53-DUC-DSNT.

When unattended substations are popular, the knob is a vital monitoring object for unattended substations. However, in the actual scene of the substation, the recognition method of a knob gear has low accuracy. The main reasons are as follows. Firstly, the SNR of knob images is low due to the influence of lighting conditions, which are challenging to extract image features. Secondly, the image deviates from the front view affected by the shooting angle; that knob has a certain deformation, which causes the feature judgment to be disturbed. Finally, the feature distribution of each kind of knob is inconsistent, which interferes with image extraction features and leads to weak spatial generalization ability. For the above problems, we propose a three-stage knob gear recognition method based on YOLOv4 and Darknet53-DUC-DSNT models for the first time and apply key point detection of deep learning to knob gear recognition for the first time. Firstly, YOLOv4 is used as the knob area detector to find knobs from a picture of a cabinet panel. Then, Darknet53, which can extract features, is used as the backbone network for keypoint detection of knobs, combined with DUC structure to recover detailed information and DSNT structure to enhance feature extraction and improve spatial generalization ability. Finally, we obtained the knob gear by calculating the angle between the line of the rotating center point and the pointing point and horizontal direction. The experimental results show that this method effectively solves the above problems and improves the performance of knob gear detection.