Deep Learning-based Artificial Intelligence On Diagnosis And Prediction Of Anterior Cruciate Ligament Injury

First part:The performance of deep learning on detection of anterior cruciate ligament lesionPurpose:Magnetic resonance imaging(MRI)is the most commonly used imaging method for diagnosing anterior cruciate ligament(ACL)injuries.However,the interpretation of knee MRI is time-intensive and depends on the clinical experience of the reader.An automated detection system based on a deep-learning algorithm may improve interpretation time and reliability.The purpose of this study was to determine the feasibility of using a deep learning approach to detect ACL injuries within the knee joint on MRI.Materials and Methods:The knee MRI data sets of 163 subjects with an ACL tear and 245 subjects with an intact ACL were retrospectively analyzed by deep learning method.The sequence is sagittal proton density-weighted spectral attenuated inversion recovery sequences at 1.5T and 3.0T.Based on the architecture of 3D DenseNet,we constructed a classification convolutional neural network for ACL injury.Using arthroscopic results as the reference standard,the performance of three different inputs and three different algorithms were compared by classification accuracy,sensitivity,specificity,positive predictive value(PPV),and negative predictive value(NPV).At the same time,we compared the performance of ACL injury detection system and radiologists.The area under receiver operating characteristic curve(AUC)was used to further evaluate the diagnostic performance of the proposed predictive model of ACL injury.Results:The accuracy,sensitivity,specificity,PPV,and NPV of our customized 3D deep learning architecture was 0.957,0.976,0.944,0.940,and 0.976,respectively.The average AUCs were 0.946,0.859,0.960 for ResNet,VGG16,and our proposed network,respectively.The diagnostic accuracy of our model,residents,and senior radiologists was 0.957,0.814,and 0.899,respectively.The customized 3D deep learning architecture demonstrates high performance in detection of ACL injuries.Conclusion:Our study demonstrated the feasibility of using an automated deeplearning-based detection system to evaluate ACL injury.The second part:Automatic segmentation model of intercondylar fossa based on deep learning:evaluating the correlation between intercondylar fossa volume and anterior cruciate ligament injuryPurpose:In this study,we propose a MRI segmentation model of intercondylar fossa based on deep learning,which is simple,fast,and automatic to measure the volume of intercondylar fossa.To explore the correlation between the volume of intercondylar fossa and ACL injury has application value in predicting and preventing ACL injury,as well as guiding the design of operation plan after ACL injury.Materials and Methods:The MRI data sets of 363 subjects with non-contact sports ACL injuries(311 male and 52 female)and 232 subjects with intact ACL(147 male and 85 female)were retrospectively analyzed.Each layer of intercondylar fossa were manually traced by radiologists on axial MRI images,then the notch volume was calculated.Unpaired t test was performed to determine the differences between the ACL-injured group and the control group,and the gender differences were also compared.Based on the architecture of Res-UNet,we constructed an automatic segmentation system for the intercondylar fossa.The quantitative index of automatic segmentation accuracy is calculated using dice similarity coefficient(DSC)to compare the performance of segmentation systems based on different networks.Results:The notch volume was extremely smaller in the ACL-injured group than that in the control group(6.64±1.47 vs 7.23±1.83 cm3,p<0.00).Females tend to have smaller notch volume than males(5.69±1.31 vs 7.23±1.57 cm3,p<0.001).The DSCs of the intercondylar fossa based on different networks were all more than 0.90,and Res-UNet showed the best performance,indicating that the automatic segmentation method proposed in this study can accurately segment the intercondylar fossa.Conclusion:The knees with smaller notch volume tended to have a higher risk of ACL injury.Using deep neural network to segment intercondylar fossa automatically provides technical support for clinical prediction and prevention of ACL injury.