Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2020
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T10:44:08.446816Z
status: published
lens: g_space
sector: health
headline: Comparison of the Performances of Deep Learning Methods in Kellgren-Lawrence Classification of Knee Osteoarthritis: A Multicenter Study
dek: Knee osteoarthritis (OA) causes symptoms such as pain, stiffness, and functional limitation in the knees; plain knee radiographs are used in addition to physical examination in its diagnosis. This study aimed to compare the performance of pre-trained deep learning models in knee OA classification and to examine the effects of lateral knee radiographs and patient clinical data on the performance of these models. Eight centers participated in the study; radiographs from six centers were used for training, validati…
gain_title: VGG-16 classified Kellgren-Lawrence knee osteoarthritis grades better than four other pretrained architectures, reaching an AUC of 0.93 and a quadratic weighted kappa of 0.91 on test data.
problem_title: (none)
trace_subject: (none)
gain_reading: VGG-16 classified Kellgren-Lawrence knee osteoarthritis grades better than four other pretrained architectures, reaching an AUC of 0.93 and a quadratic weighted kappa of 0.91 on test data.
gain_evidence: quadratic weighted Kappa of 0.91, and AUC of 0.93
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers at eight centers compared five pretrained deep learning architectures for grading knee osteoarthritis on the Kellgren-Lawrence scale. They used radiographs and clinical data from 1550 patients, training and testing on six centers and validating externally on two. VGG-16 performed best, with an accuracy of 0.73 on internal testing.

The work suggests that a single anteroposterior radiograph may be enough for automated grading, since lateral views and clinical details such as age, gender, and BMI added little. Performance on the two external centers was lower, which raises the question of how well the model would transfer to new clinical settings.
limitation: 
tag: Evidence-backed gain
key_points: Five architectures were compared: VGG-16, ResNet-50, DenseNet-121, Inception_v3, and EfficientNetb0. | VGG-16 reached a sensitivity of 0.72, a specificity of 0.93, and an F1 score of 0.72 on the main test set. | On external validation, VGG-16 had an accuracy of 0.65 and an AUC of 0.90. | Adding age, gender, and BMI features with XGBoost gave only statistically insignificant 1-2% gains. | Adding lateral radiographs to anteroposterior views gave no additional benefit for classification.
rundown: The study enrolled 1550 patients across eight centers. Six centers supplied data for training, validation, and testing, and two held out centers were used for external validation. Anteroposterior and lateral knee radiographs were collected together with age, gender, height, and weight.

VGG-16 was used for feature extraction, and the image features were combined with age, gender, and BMI and classified with machine learning algorithms. The XGBoost classifier improved metrics by only 1-2%, which was not statistically significant.

External validation results were lower than internal ones, with an accuracy of 0.65 versus 0.73 and an AUC of 0.90 versus 0.93. Quadratic weighted kappa stayed relatively high at 0.88.
sources:
- peer_reviewed | Journal of Imaging Informatics in Medicine | https://doi.org/10.1007/s10278-026-02392-2 | 2026-10-13
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

735d22763762db202ddaca6939ee3a438dc707228b85c6a821d31d438bafa2f8

previous

0000000000000000000000000000000000000000000000000000000000000000