TY - JOUR
T1 - Automatic Localization and Brand Detection of Cervical Spine Hardware on Radiographs Using Weakly Supervised Machine Learning
AU - Dutt, Raman
AU - Mendonca, Dylan
AU - Phen, Huai Ming
AU - Broida, Samuel
AU - Ghassemi, Marzyeh
AU - Gichoya, Judy
AU - Banerjee, Imon
AU - Yoon, Tim
AU - Trivedi, Hari
N1 - Publisher Copyright:
© RSNA, 2022.
PY - 2022/3
Y1 - 2022/3
N2 - Purpose: To develop an end-to-end pipeline to localize and identify cervical spine hardware brands on routine cervical spine radiographs. Materials and Methods: In this single-center retrospective study, patients who received cervical spine implants between 2014 and 2018 were identified. Information on the implant model was retrieved from the surgical notes. The dataset was filtered for implants present in at least three patients, which yielded five anterior and five posterior hardware models for classification. Images for training were manually annotated with bounding boxes for anterior and posterior hardware. An object detection model was trained and implemented to localize hardware on the remaining images. An image classification model was then trained to differentiate between five anterior and five posterior hardware models. Model performance was evaluated on a holdout test set with 1000 iterations of bootstrapping. Results: A total of 984 patients (mean age, 62 years 6 12 [standard deviation]; 525 women) were included for model training, validation, and testing. The hardware localization model achieved an intersection over union of 86.8% and an F1 score of 94.9%. For brand classification, an F1 score, sensitivity, and specificity of 98.7% 6 0.5, 98.7% 6 0.5, and 99.2% 6 0.3, respectively, were attained for anterior hardware, with values of 93.5% 6 2.0, 92.6% 6 2.0, and 96.1% 6 2.0, respectively, attained for posterior hardware. Conclusion: The developed pipeline was able to accurately localize and classify brands of hardware implants using a weakly supervised learning framework.
AB - Purpose: To develop an end-to-end pipeline to localize and identify cervical spine hardware brands on routine cervical spine radiographs. Materials and Methods: In this single-center retrospective study, patients who received cervical spine implants between 2014 and 2018 were identified. Information on the implant model was retrieved from the surgical notes. The dataset was filtered for implants present in at least three patients, which yielded five anterior and five posterior hardware models for classification. Images for training were manually annotated with bounding boxes for anterior and posterior hardware. An object detection model was trained and implemented to localize hardware on the remaining images. An image classification model was then trained to differentiate between five anterior and five posterior hardware models. Model performance was evaluated on a holdout test set with 1000 iterations of bootstrapping. Results: A total of 984 patients (mean age, 62 years 6 12 [standard deviation]; 525 women) were included for model training, validation, and testing. The hardware localization model achieved an intersection over union of 86.8% and an F1 score of 94.9%. For brand classification, an F1 score, sensitivity, and specificity of 98.7% 6 0.5, 98.7% 6 0.5, and 99.2% 6 0.3, respectively, were attained for anterior hardware, with values of 93.5% 6 2.0, 92.6% 6 2.0, and 96.1% 6 2.0, respectively, attained for posterior hardware. Conclusion: The developed pipeline was able to accurately localize and classify brands of hardware implants using a weakly supervised learning framework.
KW - Convolutional Neural Network
KW - Deep Learning Algorithms
KW - Machine Learning Algorithms
KW - Prostheses
KW - Semisupervised Learning
KW - Spine
UR - http://www.scopus.com/inward/record.url?scp=85128075406&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85128075406&partnerID=8YFLogxK
U2 - 10.1148/RYAI.210099
DO - 10.1148/RYAI.210099
M3 - Article
AN - SCOPUS:85128075406
SN - 2638-6100
VL - 4
JO - Radiology: Artificial Intelligence
JF - Radiology: Artificial Intelligence
IS - 2
M1 - e210099
ER -