Predicting Time to Total Knee Replacement Using Deep Learning
Author Information
Author(s): Ozkan Cigdem, Shengjia Chen, Chaojie Zhang, Kyunghyun Cho, Richard Kijowski, Cem Deniz
Primary Institution: New York University Grossman School of Medicine
Hypothesis
Integrating clinical variables, quantitative and semiquantitative assessments from radiographs and MRIs, and deep learning features into survival models will result in more accurate estimations of time-to-TKR compared to models using a single modality.
Conclusion
The proposed model demonstrated the potential of self-supervised learning and multimodal data fusion in accurately predicting time-to-TKR that may assist physicians to develop personalized treatment strategies.
Supporting Evidence
- The model achieved an area under the curve of 94.5 for predicting the time-to-TKR.
- Integrating self-supervised deep learning features with clinical variables improved prediction accuracy.
- The study utilized data from the Osteoarthritis Initiative and the Multicenter Osteoarthritis Study.
Takeaway
The study created a smart model that helps doctors predict when a patient might need knee surgery by looking at different types of medical images and patient information.
Methodology
The study used a survival analysis model developed from features extracted from medical images and clinical measurements, utilizing deep learning techniques.
Potential Biases
The imputation method for missing data may introduce bias, especially with nonrandom missing data.
Limitations
The study's generalizability may be limited due to the sample being primarily older, overweight, and Caucasian subjects, and the use of knees from the same patient may introduce correlation.
Participant Demographics
The study included 895 knees from the OAI dataset, with a mean age of 64.2 years for males and 62.4 years for females.
Statistical Information
P-Value
<.001
Confidence Interval
94.0-95.1
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website