Enhanced deep learning model enables accurate alignment measurement across diverse institutional imaging protocols
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
Background: Achieving consistent accuracy in radiographic measurements across different equipment and protocols is challenging. This study evaluates an advanced deep learning (DL) model, building upon a precursor, for its proficiency in generating uniform and precise alignment measurements in full-leg radiographs irrespective of institutional imaging differences. Methods: The enhanced DL model was trained on over 10,000 radiographs. Utilizing a segmented approach, it separately identified and evaluated regions of interest (ROIs) for the hip, knee, and ankle, subsequently integrating these regions. For external validation, 300 datasets from three distinct institutes with varied imaging protocols and equipment were employed. The study measured seven radiologic parameters: hip-knee-ankle angle, lateral distal femoral angle, medial proximal tibial angle, joint line convergence angle, weight-bearing line ratio, joint line obliquity angle, and lateral distal tibial angle. Measurements by the model were compared with an orthopedic specialist's evaluations using inter-observer and intra-observer intraclass correlation coefficients (ICCs). Additionally, the absolute error percentage in alignment measurements was assessed, and the processing duration for radiograph evaluation was recorded. Results: The DL model exhibited excellent performance, achieving an inter-observer ICC between 0.936 and 0.997, on par with an orthopedic specialist, and an intra-observer ICC of 1.000. The model's consistency was robust across different institutional imaging protocols. Its accuracy was particularly notable in measuring the hip-knee ankle angle, with no instances of absolute error exceeding 1.5 degrees. The enhanced model significantly improved processing speed, reducing the time by 30-fold from an initial 10-11 seconds to 300 milliseconds. Conclusions: The enhanced DL model demonstrated its ability for accurate, rapid alignment measurements in full-leg radiographs, regardless of protocol variations, signifying its potential for broad clinical and research applicability.
Abstract
Background: Achieving consistent accuracy in radiographic measurements across different equipment and protocols is challenging. This study evaluates an advanced deep learning (DL) model, building upon a precursor, for its proficiency in generating uniform and precise alignment measurements in full-leg radiographs irrespective of institutional imaging differences. Methods: The enhanced DL model was trained on over 10,000 radiographs. Utilizing a segmented approach, it separately identified and evaluated regions of interest (ROIs) for the hip, knee, and ankle, subsequently integrating these regions. For external validation, 300 datasets from three distinct institutes with varied imaging protocols and equipment were employed. The study measured seven radiologic parameters: hip-knee-ankle angle, lateral distal femoral angle, medial proximal tibial angle, joint line convergence angle, weight-bearing line ratio, joint line obliquity angle, and lateral distal tibial angle. Measurements by the model were compared with an orthopedic specialist's evaluations using inter-observer and intra-observer intraclass correlation coefficients (ICCs). Additionally, the absolute error percentage in alignment measurements was assessed, and the processing duration for radiograph evaluation was recorded. Results: The DL model exhibited excellent performance, achieving an inter-observer ICC between 0.936 and 0.997, on par with an orthopedic specialist, and an intra-observer ICC of 1.000. The model's consistency was robust across different institutional imaging protocols. Its accuracy was particularly notable in measuring the hip-knee ankle angle, with no instances of absolute error exceeding 1.5 degrees. The enhanced model significantly improved processing speed, reducing the time by 30-fold from an initial 10-11 seconds to 300 milliseconds. Conclusions: The enhanced DL model demonstrated its ability for accurate, rapid alignment measurements in full-leg radiographs, regardless of protocol variations, signifying its potential for broad clinical and research applicability.
참고문헌 (29)
- Babazadeh S, "The long leg radiograph is a reliable method of assessing alignment when compared to computer-assisted navigation and computer tomography", Knee, 20(4), pp.242~249 (2013)[DOI]
- Skyttä ET, "Comparison of standard anteroposterior knee and hip-to-ankle radiographs in determining the lower limb and implant alignment after total knee arthroplasty", Scand J Surg, 98(4), pp.250~253 (2009)[DOI]
- Federer SJ, "Artificial intelligence in orthopaedics : a scoping review", PLoS ONE, 16(11), pp.e0260471~ (2021)[DOI]
- Simon S, "Fully automated deep learning for knee alignment assessment in lower extremity radiographs : a cross-sectional diagnostic study", Skeletal Radiol, 51(6), pp.1249~1259 (2022)[DOI]
- Tack A, "Fully automated Assessment of Knee Alignment from Full-Leg X-Rays employing a"YOLOv4 And Resnet Landmark regression Algorithm"(YARLA) : Data from the Osteoarthritis Initiative", Comput Methods Programs Biomed, 205, pp.106080~ (2021)[DOI]
- Erne F, "Automated artificial intelligence-based assessment of lower limb alignment validated on weight-bearing pre- and postoperative full-leg radiographs", Diagnostics (Basel) (2022)[DOI]
- Jo C, "Deep learning-based landmark recognition and angle measurement of full-leg plain radiographs can be adopted to assess lower extremity alignment", Knee Surg Sports Traumatol Arthrosc, 31(4), pp.1388~1397 (2023)[DOI]
- Moon KR, "A deep learning approach for fully automated measurements of lower extremity alignment in radiographic images", Sci Rep, 13(1), pp.14692~ (2023)[DOI]
- Nguyen TP, "Intelligent analysis of coronal alignment in lower limbs based on radiographic image with convolutional neural network", Comput Biol Med, 120, pp.103732~ (2020)[DOI]
- Paley D, "Mechanical axis deviation of the lower limbs. Preoperative planning of uniapical angular deformities of the tibia or femur", Clin Orthop Relat Res, 280, pp.48~64 (1992)
- Bellemans J, "The Chitranjan Ranawat award: is neutral mechanical alignment normal for all patients? The concept of constitutional varus", Clin Orthop Relat Res, 470(1), pp.45~53 (2012)[DOI]
- Tseng TH, "Knee-ankle joint line angle : a significant contributor to high-degree knee joint line obliquity in medial opening wedge high tibial osteotomy", J Orthop Surg Res, 17(1), pp.79~ (2022)[DOI]
- Ji HM, "Kinematically aligned TKA can align knee joint line to horizontal", Knee Surg Sports Traumatol Arthrosc, 24(8), pp.2436~2441 (2016)[DOI]
- McArthur JR, "Resemblance of valgus malalignment of the distal tibia in 15-degree craniocaudal radiographs", J Orthop Surg(Hong Kong), 21(3), pp.337~339 (2013)[DOI]
- Shorten C, "A survey on image data augmentation for deep learning", J Big Data, 6(1), pp.1~48 (2019)
- Geng ZG, "Bottom-up human pose estimation via disentangled keypoint regression", Proc Cvpr IEEE (2021)
- Wang XY, "Adaptive wing loss for robust face alignment via heatmap regression", IEEE I Conf Comp Vis, 10 (2019)
- Gielis WP, "Predicting the mechanical hip-knee-ankle angle accurately from standard knee radiographs : a cross-validation experiment in 100 patients", Acta Orthop, 91(6), pp.732~737 (2020)[DOI]
- Eng J, "Getting started in radiology research : asking the right question and identifying an appropriate study population", Acad Radiol, 11(2), pp.149~154 (2004)[DOI]
- Blackmore CC, "The challenge of clinical radiology research", AJR Am J Roentgenol, 176(2), pp.327~331 (2001)[DOI]
- Yu AC, "External validation of deep learning algorithms for radiologic diagnosis : a systematic review", Radiol Artif Intell, 4(3), pp.e210064~ (2022)[DOI]
- Na YG, "Change of joint-line convergence angle should be considered for accurate alignment correction in high tibial osteotomy", Knee Surg Relat Res, 33(1), pp.4~ (2021)[DOI]
- Hodel S, "A novel approach for joint line restoration in revision total ankle arthroplasty based on the three-dimensional registration of the contralateral tibia and fibula", J Exp Orthop, 10(1), pp.10~ (2023)[DOI]
- Lin DY, "RefineU-Net : Improved U-Net with progressive global feedbacks and residual attention guided local refinement for medical image segmentation", Pattern Recogn Lett, 138, pp.267~275 (2020)[DOI]
- Artacho B, "Full-BAPose : bottom up framework for full body pose estimation", Sensors-Basel, 23(7), pp.3725~ (2023)
- Wang JD, "Deep high-resolution representation learning for visual recognition", IEEE T Pattern Anal, 43(10), pp.3349~3364 (2021)[DOI]
- Sun K, "Wang JD(2019)Deep high-resolution representation learning for human pose estimation", 2019 IEEE, 2019, 5686~5696
- Gong ML, "A review of non-maximum suppression algorithms for deep learning target detection", Proc Spie, pp.11763~ (2021)
- Hosang J, "Schiele B(2017)Learning non-maximum suppression", 30th IEEE Conference on Computer Vision and Pattern Recognition (Cvpr 2017), pp.6469~6477 (2017)