Classification Diagnosis Of Knee Cartilage Injury Based On T2-Mapping Convolutional Neural Network

PurposeTo construct a AlexNet Architecture base on MRI conventional sequence images and T2-mapping images,achieve the classification diagnosis of knee cartilage injury.Materials and methods1.Clinical data131 patients diagnosed with knee cartilage injury and underwent arthroscopic surgery from June 2021 to January 2023 from the Second Affiliated Hospital of Dalian Medical University and Dalian Third People’s Hospital were collected,The patients,aged 18-55 years old,according to arthroscopy as the gold standard and ICRS(International Cartilage Repair Society)cartilage damage classification,were classified as grade Ⅰ 30,Grade Ⅱ 32,grade Ⅲ 35,and Grade Ⅳ 34.25 healthy volunteers were chose,without knee discomfort and surgical history.Inclusion criteria:(1)Patients aged 18-55 years old.(2)Patients with knee pain or mobility disorder were seen for 2-3 months.Exclusion criteria:(1)high-risk groups over 55 years old or with knee joint degeneration;(2)Patients with previous knee lesions or knee surgery history(3)contraindicated MRI examination.2.MRI equipment and methodSiemens skyra3.0T magnetic resonance scanner and 15-channel Knee joint coil was used for examination.All healthy volunteers and patients underwent routine knee joint magnetic resonance scanning(TiWI、PDWI)and T2-Mapping(TR 1830ms,TE 13.8 ms,FOV 170mm ×170 mm,matrix 384×384,layer thickness 3.0 mm,FA 180°,average acquisition times 1)sequence examination.3.Image processingThree methods included manual diagnosis,SqueezeNet Architecture and AlexNet Architecture diagnosis the grade of knee cartilage injury.The MR image of cartilage injury was evaluated by an experienced radiologist.The cartilage injury was graded by two doctors first.If the results were inconsistent,the three doctors would make grading diagnosis together.For different levels of damage at the same location,the highest damage level is the result.The above two models first import the original image and data enhancement image into the model as data sets;Secondly,the model is trained through the data set and the model parameters are continuously optimized;Then,randomly select 30%of the images in the data set as the test set and input them into the trained model;Finally,the predicted classification results are obtained.4.Statistical analysisMatLab software and SPSS 26.0 statistical analysis software are used.Calculate the accuracy,recall rate,F1-score and accuracy rate of different methods for diagnosis of knee cartilage injury grading;use Chi-square Test and compare.P<0.05 is statistically significant.Results1.Manual diagnosisNormal knee joint with precision(90.91%)、recall(86.96%)、F1-score(88.89%);Knee cartilage damage diagnostic value of Ⅰ grade with precision(68.18%)、recall(65.21%)、F1score(66.64%);Knee cartilage damage diagnostic value of Ⅱ grade with precision(79.17%)、recall(76.00%)、F1-score(77.33%);Knee cartilage damage diagnostic value of Ⅲ grade with precision(88.46%)、recall(85.19%)、F1-score(86.80%);Knee cartilage damage diagnostic value of Ⅳ grade with precision(91.67%)、recall(88.00%)、F1-score(89.80%).2.SqueezeNet ArchitectureNormal knee joint with precision(98.65%)、recall(97.33%)、F1-score(97.99%);Knee cartilage damage diagnostic value of Ⅰ grade with precision(97.05%)、recall(98.67%)、F1score(97.85%);Knee cartilage damage diagnostic value of Ⅱ grade with precision(96.25%)、recall(94.00%)、F1-score(95.11%);Knee cartilage damage diagnostic value of Ⅲ grade with precision(92.76%)、recall(94.00%)、F1-score(93.38%);Knee cartilage damage diagnostic value of Ⅳ grade with precision(99.34%)、recall(100.00%)、F1-score(99.67%).3.AlexNet ArchitectureNormal knee joint with precision(99.33%)、recall(99.00%)、F1-score(99.17%);Knee cartilage damage diagnostic value of Ⅰ grade with precision(99.66%)、recall(98.67%)、F1score(99.16%);Knee cartilage damage diagnostic value of Ⅱ grade with precision(94.70%)、recall(95.33%)、F1-score(95.02%);Knee cartilage damage diagnostic value of Ⅲ grade with precision(95.61%)、recall(94.33%)、F1-score(94.97%);Knee cartilage damage diagnostic value of Ⅳ grade with precision(98.04%)、recall(100.00%)、F1-score(99.01%).Conclusion1.AlexNet and SqueezeNet classification models based on conventional MRI sequence images and T2-Mapping image,has better performance in the diagnosis of knee cartilage injury than Manual diagnosis.2.AlexNet,a classification model based on conventional magnetic resonance sequence images and T2-Mapping images,has better performance in the classification diagnosis of knee cartilage injury than SqueezeNet.3.AlexNet,a classification model based on conventional MRI sequence images and T2-Mapping image,has more prominent advantages in the diagnosis of grade Ⅰ and Ⅱknee injury.