Automated classification of chondroid tumor using 3D U-Net and radiomics with deep features

Abstract

Classifying chondroid tumors is an essential step for effective treatment planning. Recently, with the advances in computer-aided diagnosis and the increasing availability of medical imaging data, automated tumor classification using deep learning shows promise in assisting clinical decision-making. In this study, we propose a hybrid approach that integrates deep learning and radiomics for chondroid tumor classification. First, we performed tumor segmentation using the nnUNetv2 framework, which provided three-dimensional (3D) delineation of tumor regions of interest (ROIs). From these ROIs, we extracted a set of radiomics features and deep learning-derived features. After feature selection, we identified 15 radiomics and 15 deep features to build classification models. We developed 5 machine learning classifiers including Random Forest, XGBoost, Gradient Boosting, LightGBM, and CatBoost for the classification models. The approach integrating features from radiomics, ROI-originated deep learning features, and clinical variables yielded the best overall classification results. Among the classifiers, CatBoost classifier achieved the highest accuracy of 0.90 (95% CI 0.90-0.93), a weighted kappa of 0.85, and an AUC of 0.91. These findings highlight the potential of integrating 3D U-Net-assisted segmentation with radiomics and deep learning features to improve classification of chondroid tumors.

Keywords: 3D U-Net; Chondroid tumors; Computer-aided diagnosis; Deep learning; MRI images; Machine learning algorithms; Radiomics; Tumor segmentation.

Fig. 1

Comparison of ROC curve between models on chondroid tumor classification. The curve plots the true positive rate against the false positive rate across five models: Model 1 (Radiomics-Only, blue), Model 2 (Deep Learning-Only, green), Model 3 (Radiomics + Deep Learning, pink), Model 4 (Radiomics + Deep Learning + Clinical, red), and Model 5 (Radiomics + Deep ROI + Clinical, cyan). The dashed line illustrated the random classifier (AUC = 0.5). Model 5 shows the highest AUC, indicating the best classification performance, followed by Model 4, 3, 2, and 1.

Fig. 2

Confusion Matrix of CatBoost classifier across different combined-features datasets. Each matrix represents performance across 5 models: (Top left) Model 1: Radiomics-Only, (Top right) Model 2: Deep Learning-Only, (Middle left) Model 3: Radiomics + Deep Learning, (Middle right) Model 4: Radiomics + Deep Learning + Clinical, (Bottom left) Model 5: Radiomics + Deep ROI + Clinical. These matrices plot a number of correct and incorrect samples in each class (Class 0: Enchondroma, Class 1: Grade 1 chondrosarcoma, and Class 2: Grade 2 & 3 chondrosarcoma). Darker cells along the diagonal indicate a larger number of correct classification counts, suggesting the enhanced performance of feature integration models, particularly Model 5.

Fig. 3

Patient selection flowchart for chondroid tumor classification based on MRI sequences from 2008 to 2021.

Fig. 4

Illustration of 3D U-Net architecture for chondroid tumor segmentation. This model works on 3D image patches size of [64 × 64 × 64] with voxel spacing standardized to [2.0, 0.265, 0.265] mm. The five-stage decoder-encoder structure reduces the resolution to half in each encoder stage, starting with the input of (64 × 64 × 64), and then decreasing to (32 × 32 × 32), (16 × 16 × 16), (8 × 8 × 8), and finally (4 × 4 × 4). The decoder path restores the resolution by upsampling, reversing the process to the original dimension. The skip connections link each encoder layer to the corresponding decoder layer to retain spatial information.

Fig. 5

Classification pipeline incorporating radiomics, deep learning, deep ROI, and clinical features to improve performance.