Related Experiment Video
Updated: Aug 6, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
REST Framework in Action: Predicting Benign versus Malignant Bone Tumors-An External Validation Study
Duddukunta Vishal Reddy1, Ch Raja Bhaskar Venkata Sri Madhusudan1, Manish Kumar Shah1
1Department of Orthopedic Surgery, AIIMS (All India Institute of Medical Sciences), Shivaji Nagar, Rishikesh, Uttarakhand 249203 India.
Objective:
The Radiological Evaluation Score for Bone Tumors (REST) system is a radiographic tool of excellent diagnostic accuracy with a score of >3, being 100% predictive of malignant bone lesions. The present study is designed to externally validate the REST tool among reviewers of varying clinical experience and against their clinical gestalt.
Methodology:
The REST system was externally validated on the radiographs of primary bone tumors with a confirmed histopathological diagnosis. Three reviewers (MDH, HRS, MSH) independently reviewed 104 radiographs of primary bone tumors using the REST system to provide cumulative and individual radiographic factor score. Three separate reviewers (ARJ, RHL, SCH) independently reviewed radiographs based on their clinical gestalt and rated them as either benign or malignant bone lesions. The institutional histopathological report was considered the reference standard for the validation. The characteristics of diagnostic accuracy within and between the groups were measured by sensitivity and specificity and receiver operator characteristic curves. The inter-group and intra-group observer correlation was calculated with 95% confidence intervals.
Results:
104 radiographs of bone lesions with equal incidence of benign and malignant lesions were included in this study. The sensitivity and specificity of the REST Score group for diagnosing a malignant bone lesion were 80.8% and 84.6%, whereas the sensitivity and specificity of the clinical gestalt group were 63.5% and 90.4%, and this difference was statistically significant (p = 0.017). However, the difference in area under the curve 0.058 (-0.028 to 0.143) between the REST score group and clinical gestalt group, however, was not statistically significant p = 0.187. The reviewers in the REST group had greater interobserver reliability when compared to that of the clinical gestalt group [0.738 (0.637-0.815) vs. 0.446 (0.336-0.557)].
Conclusion:
The REST is a radiographic tool of considerable diagnostic accuracy that can be used as a key step in the systematic evaluation of bone lesions.