Automated evaluation of rheumatoid arthritis from hand radiographs using Machine Learning and deep learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Rheumatoid Arthritis (RA) diagnosis relies on interpreting hand radiographs.
• Automated analysis of medical images can improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a customized Convolutional Neural Network (CNN) model for automated classification of hand X-ray images.
• To compare the performance of the custom CNN model against modified pre-trained CNNs and traditional machine learning classifiers.
• To investigate the efficacy of fusing handcrafted features (SIFT) with CNN-extracted features for RA detection.

Main Methods:

• Implementation of automated patch-based classification using modified pre-trained CNNs (GoogLeNet) and a custom-developed CNN model.
• Training and testing the models on a dataset of hand radiographs, utilizing 10,000 patches from 75 images for training and 500 patches from 25 images for testing.
• Feature fusion of SIFT and custom CNN features, followed by classification using Machine Learning (ML) classifiers.

Main Results:

• The custom CNN model (custom3) achieved a high accuracy of 95%, surpassing the modified GoogLeNet model's accuracy of 89%.
• The custom3 model demonstrated superior sensitivity (95%) and specificity (94%) compared to GoogLeNet (84% sensitivity, 90% specificity).
• Feature fusion (SIFT + CNN) with the custom3 model showed better performance in classifying RA compared to ML classifiers using only SIFT or CNN features.

Conclusions:

• A custom CNN-based approach provides an effective computer-aided diagnostic tool for detecting Rheumatoid Arthritis from hand X-rays.
• The developed custom CNN model significantly outperforms existing pre-trained models and traditional methods in terms of accuracy, sensitivity, and specificity.
• Automated feature extraction and classification using customized CNNs hold great promise for enhancing the diagnostic process in rheumatology.