# Nomograms Based on X-Ray Radiomics for Predicting Pain Progression in Knee Osteoarthritis Using Data From the Foundation for the National Institutes of Health: Development and Validation Study

## TL;DR

This study developed and validated a radiomics-based nomogram to predict knee osteoarthritis pain progression using X-ray images and clinical data.

## Contribution

The novel contribution is the creation of interpretable radiomics nomograms for predicting KOA pain progression using automated image analysis and clinical indicators.

## Key findings

- Wavelet-HH_gldm_HighGrayLevelEmphasis was identified as the primary radiomics feature for predicting pain progression.
- Nomograms achieved area under the curve values of 0.766 and 0.753 with strong calibration and decision curve benefits.
- Subgroup analyses showed areas under the curve of 0.795 and 0.740 for pain progression prediction.

## Abstract

Knee osteoarthritis (KOA) is one of the most prevalent chronic musculoskeletal disorders among the older adult population. Screening populations at risk of rapid progression of osteoarthritis and implementing appropriate early intervention strategies is advantageous for the treatment and prognosis of affected patients.

This study aimed to construct and validate a nomogram model based on x-ray radiomics to effectively identify individuals experiencing progression of KOA pain.

The Foundation for the National Institutes of Health Biomarkers Consortium included a total of 600 participants who were classified as pain progressors (n=297, 49.5%) and non–pain progressors (n=303, 50.5%) according to an increase in the Western Ontario and McMaster Universities Osteoarthritis Index pain score of ≥9 points (on a scale from 0 to 100) during the follow-up period of 24 to 48 months. X-rays that lacked defined spacing in the DICOM image were excluded. Fully automatic selection of subchondral bone regions on the inner and outer edges of the tibia and femur as regions of interest and extraction of radiomics features for different combinations of regions of interest were conducted. Least absolute shrinkage and selection operator regression was used to select features and generate a radiomics score using Shapley additive explanations for interpretability. The radiomics score, along with clinical indicators, was incorporated into nomograms using a multivariable logistic regression model. The subgroup analysis focused solely on the progression of pain and cases with no progression at all. The receiver operating characteristic curve, along with calibration and decision curves, was used to assess the discriminative performance.

A total of 450 participants were included in the study. Shapley additive explanations analysis identified Wavelet-HH_gldm_HighGrayLevelEmphasis as the primary radiomics feature. Nomogram 1 and nomogram 2 for predicting KOA pain progression achieved area under the curve values of 0.766 and 0.753, respectively, with mean absolute errors of 0.012 and 0.008, respectively, in the calibration curves. Decision curve analysis showed a positive net benefit across a range of threshold probabilities. In subgroup analyses, nomogram 3 and nomogram 4 yielded areas under the curve of 0.795 and 0.740, respectively.

The nomograms based on x-ray radiomics demonstrated excellent predictive capability and accuracy in forecasting the progression of KOA pain.