Radiomics and Machine Learning Can Differentiate Transient Osteoporosis from Avascular Necrosis of the Hip

Abstract

Differentiation between transient osteoporosis (TOH) and avascular necrosis (AVN) of the hip is a longstanding challenge in musculoskeletal radiology. The purpose of this study was to utilize MRI-based radiomics and machine learning (ML) for accurate differentiation between the two entities. A total of 109 hips with TOH and 104 hips with AVN were retrospectively included. Femoral heads and necks with segmented radiomics features were extracted. Three ML classifiers (XGboost, CatBoost and SVM) using 38 relevant radiomics features were trained on 70% and validated on 30% of the dataset. ML performance was compared to two musculoskeletal radiologists, a general radiologist and two radiology residents. XGboost achieved the best performance with an area under the curve (AUC) of 93.7% (95% CI from 87.7 to 99.8%) among ML models. MSK radiologists achieved an AUC of 90.6% (95% CI from 86.7% to 94.5%) and 88.3% (95% CI from 84% to 92.7%), respectively, similar to residents. The general radiologist achieved an AUC of 84.5% (95% CI from 80% to 89%), significantly lower than of XGboost (p = 0.017). In conclusion, radiomics-based ML achieved a performance similar to MSK radiologists and significantly higher compared to general radiologists in differentiating between TOH and AVN.

Keywords: XGboost; artificial intelligence; avascular necrosis of bone; hip; machine learning; osteoporosis; radiomics; transient osteoporosis.

Figure 1

Flow diagram describing the formation of study groups for radiomics analysis and machine learning model development. TOH: Transient Osteoporosis of the Hip; AVN: Avascular Necrosis; ML: Machine Learning (created with BioRender.com, date last accessed: 15 September 2021).

Figure 2

Computational pipeline for radiomics analysis and machine learning model development. The process starts with image acquisition and segmentation of the femoral head and neck (1) followed by radiomics analysis (2) consisting of feature extraction and data preprocessing in preparation for subsequent model development (3). Three machine learning algorithms (XGboost, CatBoost and SVM) were trained and validated with multivendor data and their performance was compared to that of expert readers. TOH: Transient Osteoporosis of the Hip; AVN: Avascular Necrosis; STIR: Short Tau Inversion Recovery; LoG: Laplacian of Gaussian; SVM: Support Vector Machine (created with BioRender.com, date last accessed: 15 September 2021).

Figure 3

Identification of important features with the use of Boruta feature selection. Following collinearity correction and scaling, Boruta was applied as an artificial intelligence algorithm to select relevant features for unbiased development of machine learning classifiers. The Z-score boxplot presents rejected (red), tentative (yellow) and accepted (green) features. p < 0.01 was used as a cutoff for the selection of accepted features. Blue boxes represent Z-scores of shadow features acting as internal controls for the selection of important variables. Subsequent machine learning was performed using accepted (green) features.

Figure 4

Receiver operating characteristic (ROC) curves of machine learning models. XGboost (A), CatBoost (B) and support vector machine (SVM) (C). Light blue areas represent the respective 95% confidence intervals calculated with bootstrapping. AUC: Area Under the Curve.

Figure 5

Radiomics features identified as important for the performance of XGboost. Important features belong to two clusters based on their degree of importance. Cluster 2 contains three features which represent the most important determinants of XGboost performance in differentiating between TOH and AVN.

Figure 6

Comparison between receiver operating characteristic (ROC) curves of XGboost and expert readers. ROC curves of XGboost and musculoskeletal radiologists are plotted as solid lines whereas the ROC curves of residents and the general radiologist are plotted as dashed lines. XGboost (pink line) is shown to have the best performance, which was significantly higher than the performance of a general radiologist (GR—purple line). XGB: XGboost; MSKR: Musculoskeletal Radiologist; GR: General Radiologist; RR: Radiology Resident; OBS: Observer; AUC: Area Under the Curve.