SLAP Injury of the Shoulder Joint: Application Value of Deep Learning in Diagnosis

SLAP Injury of the Shoulder Joint: Application Value of Deep Learning in Diagnosis

This study intends to study the shoulder SLAP injury through deep learning technology and establish a deep learning model through the combination of axial and oblique coronal images to establish a deep learning method that can accurately identify and grade shoulder SLAP injury.

1. Recognition of labrum images based on LeNet: axial and oblique coronal T2-fs images were used, and all images were corrected and standardized. LeNet identified the images with labrum of the shoulder joint, and the images with labrum structure of shoulder joint were selected from the complete sequence. In contrast, the images without labrum structure were deleted. All the data are divided into a training set (70%, 30% in training set as verification set), and the remaining 30% as a test set to evaluate the accuracy of model recognition. Enter the obtained results into the next step.
2. Recognition and segmentation of glenoid lip of shoulder joint based on DenseNet: the labrum is recognized by DenseNet in the selected image. The labelimg software based on Python was used to locate the labrum coordinates and then input them into Python for recognition learning. All the data were divided into a training set (70% and 30% of the training set were selected as the verification set). The remaining 30% was used as the test set to evaluate the accuracy of model recognition. After identifying the labrum structure, the labrum structure is locally cut and enlarged to remove the redundant information and improve the recognition efficiency and accuracy. Finally, input the result to the next step.
3. Recognition and grading of shoulder SLAP injury based on 3D-CNN: recognition and grading of input data through 3D-CNN model. 3D-CNN is divided into eight layers: input layer, hard wire layer H1, convolution layer C2, downsampling layer S3, convolution layer C4, downsampling layer S5, convolution layer C6 and output layer. 3D-CNN constructs a cube by stacking multiple consecutive frames and then uses a 3D convolution kernel in the cube. Through this structure, the feature images in the convolution layer will be connected with multiple adjacent frames in the previous layer to realize the information acquisition of continuous images. Similarly, the data is divided into a training set (70%, and then 30% of the training set is selected as the verification set), and the remaining 30% is used as the test set to evaluate the classification accuracy to identify whether there is labrum injury and grade the image with injury.
4. Establish CNN combined model: after establishing the model for the axial and oblique coronal view according to the above process (1-3), according to the output characteristics of the CNN classification model, predict the probability of different grades before the output results, and the output results are based on these probabilities to select the expression form of the maximum possible probability. Our combined model averages the probabilities of these different classifications, calculates the final prediction probability, and then obtains the final joint model. The test set of the third step (including the mixed data of axial and coronal images) was used to verify the joint model.

Inclusion Criteria:

1. Without any treatment before imaging examination;
2. MR of the shoulder joint was performed within 3 months before the operation and the image quality was good;
3. Arthroscopic operation was performed in our hospital, and the operation records were complete.

Exclusion Criteria:

1. History of shoulder surgery, tumor, or previous fracture;
2. Unclear image, serious artifact, or incomplete clinical data.