Evaluation Of Knee Cartilage Based On MRI AI Reconstruction Model Of Knee Joint

Knee osteoarthritis(KOA)is one of the most common chronic osteoarthritis.The main manifestations are joint pain,limited activity,and even deformity,which seriously affect people’s normal daily life.At present,among the medical images that can be used for clinical diagnosis,MRI is an imaging examination method to distinguish the tissues such as ligaments,synovium,cartilage,and joint fluid,which provides an inestimable value for the diagnosis and treatment of various osteoarthritis.In recent years,deep learning technology has attracted a lot of attention in the field of medical imaging,and has been successfully used in MRI image segmentation of knee joints.It provides a new method for the diagnosis,treatment and prognosis of knee arthropathy,and breaks the original mode of diagnosis made by doctors reviewing images.Objective In this study,the feasibility of the AI reconstruction model based on the thin MRI data of the knee joint to evaluate the knee joint cartilage injury was discussed by comparing the real object of the knee joint tibial platform obtained during the operation;The tibial platform was scanned by CT,and the CT-based cartilage model was reconstructed.The accuracy of the MRI based AI reconstruction of the tibial cartilage model was verified by the actual tibial platform and the CT reconstruction model.Methods We selected thirty three patients(a total of forty one knees)who were hospitalized with severe knee osteoarthritis in Beijing Tsinghua Changgung Hospital from May 2021 to April 2022.All of them were planned to be performed total knee arthroplasty(TKA)for the treatment of knee osteoarthritis.Fifteen males with an average age of 71±5 years old and twenty six females with an average age of 71±9 years old were included in this study.At the same time,there were 19 cases of left knee and 22 cases of right knee in these knees.We performed thin layer MRI examination on the patients’ knee joints before the surgery and we reconstructed artificial intelligence model based on the thin layer MRI data of the knee joint.For subsequent verification,the cartilage part of the model was selected and corrected by Principal Component Analysis(PCA)in order to realize model straightening.After the total knee arthroplasty(TKA)operation,we washed the tibial plateau of the knee joint amputated during the operation with a pulse washing gun to remove excess oil and residual soft tissue.Wipe the tibial plateau clean and place it horizontally at 1mm × 1mm grid paper,fix the camera(fixing the shooting angle,distance,pixel and other information)vertically above the tibial cartilage with the shooting bracket,and take the front image of the tibial platform.The cartilage of the medial and lateral tibial plateau was divided into three equal parts in the anteroposterior and left and right diameters to form a "well" grid division,and then the cartilage of the tibial plateau was graded and recorded according to the International Cartilage Repair Society ICRS.Third: CT scan the intercepted tibial plateau(parameter: CBCT,stadium voltage 120 k Vp,160 m As,CT reconstruction image pixel is isotropic 0.2mm),and carry out the scanning original data in two software in the personal computer: MITKWorkbench software initially labeled cartilage,and then use ITK-SNAP software to accurately label the preliminary labeling results to establish a CT-based tibial plateau cartilage model.Fourth: Compare the actual cartilage specimen of the tibial plateau with the ICRS grading results of knee joint AI reconstruction model and artificial recognition of knee joint MRI image,calculate the corresponding sensitivity,specificity,PPV,NPV and Kappa coefficient,draw the ROC curve,and calculate the area under the ROC curve(AUC).The thickness of cartilage in each region of the ICRS classification of the tibial plateau cartilage model based on thin MRI of the knee joint and the real tibial plateau cartilage specimen was measured.Fifthly,calculate the two-dimensional Dice similarity coefficient(DSC)between the physical specimen of the tibial platform and the AI reconstruction cartilage model based on the thin MRI of the knee joint,and the two-dimensional Dice similarity coefficient between the physical specimen of the tibial platform and the CT-based tibial platform cartilage model;The three-dimensional Dice similarity coefficient(DSC)between the CT-based tibial plateau cartilage model and the AI reconstruction cartilage model based on the thin MRI of the knee joint,and the cartilage volume of the CT-based cartilage model and the AI intelligent reconstruction model based on MRI were calculated.Results After the above experiments,compared with the grade of cartilage injury intercepted during our operation which is according to the ICRS classification,the sensitivity of artificial intelligence reconstruction model for the diagnosis of cartilage injury with ICRS classification grade four was 93.1 percent.The specificity of artificial intelligence reconstruction model was 91.4 percent.The positive predictive value(PPV)of artificial intelligence reconstruction model was 92.2 percent.And the negative predictive value(NPV)of artificial intelligence reconstruction model was80.3 percent.The area under ROC curve(AUC)is 0.922.The ICRS classification consistency between artificial intelligence model and physical inspection results is good which kappa value is 0.806(P <0.001).In the aspect of artificial recognition of cartilage injury grading in MRI images,the sensitivity of artificial recognition is 92.10 percent compared with the manual identification of cartilage injury classification in MRI images.The specificity of artificial recognition is 91.60 percent.The positive predictive value(PPV)of artificial recognition is 97.20 percent and the negative predictive value(NPV)of artificial recognition is 78.8 percent.The kappa value of the cartilage injury classification in MRI images consistency between artificial recognition and manual identification is 0.794(P<0.001).The average two-dimensional DSC between AI reconstruction of tibial plateau cartilage model based on thin MRI of knee joint and the actual tibial plateau cartilage sample is 0.83;Based on CT,the average two-dimensional DSC between the tibial plateau cartilage model and the real tibial plateau cartilage sample is 0.90;The three-dimensional Dice similarity coefficient(DSC)between the CT-based tibial plateau cartilage model and the AI reconstruction cartilage model based on thin MRI of the knee joint is 0.52.The Pearson correlation between the cartilage volume of artificial recognition based on CT model and that of AI reconstruction model was0.725(P<0.05).The two-dimensional DSC between AI reconstruction of tibial plateau cartilage model based on thin MRI of knee joint and the real tibial plateau cartilage sample is 0.83;Based on CT,the two-dimensional DSC between the tibial plateau cartilage model and the real tibial plateau cartilage sample is 0.90;The three-dimensional Dice similarity coefficient(DSC)between the CT-based tibial plateau cartilage model and the AI reconstruction cartilage model based on thin MRI of the knee joint is 0.52.The Pearson correlation between the cartilage volume of artificial recognition based on CT model and that of AI reconstruction model was 0.725(P<0.05).Conclusion The AI reconstruction model based on thin MRI of knee joint has good sensitivity and specificity for the diagnosis of ICRS grade IV cartilage injury;Moreover,the AI intelligent reconstruction cartilage model based on the thin MRI of the knee joint can better reflect the shape of the actual tibial plateau cartilage and the location of the cartilage defect with ICRS grade 4.However,in terms of cartilage thickness and threedimensional accuracy,the cartilage model based on MRI underestimated the actual volume of cartilage.Compared with the traditional method of evaluating the accuracy of artificial intelligence,this paper uses a new evaluation index to confirm the reliability of artificial intelligence in the current knee joint cartilage injury,and provides a verification basis and development direction for further clinical application of knee joint reconstruction model based on MR data.