Machine Learning Approach for Detection and Classification of Scoliosis in Human Bones from X-Ray Images

Abstract

Scoliosis is a prevalent orthopedic condition that affects individuals across various age groups. It is a lateral curvature of the spine and its early detection is crucial for effective treatment of the condition. This study presented a machine learning based model for detection and classification of scoliosis from X-ray images. A total of 261 X-ray images depicting the spine in posterior-anterior and lateral views which consisted of 189 positive scoliosis cases and 72 negative cases forming datasets for the study were curated from Yaba Gbobi hospital and online zip extractor. Data analysis via the use of descriptive statistics such as mean, standard deviation, minimum and maximum pixel intensities was done to extract the image properties. Data pre- processing involved resizing of the images to a standardized dimension of 224 mm x 224 mm was carried out. Augmentation of the dataset was done using Fast Artificial Intelligence (fast AI) codes written in Python Programming Language, by setting up image data loaders from folders to path to dataset which trained and validated the images selected in batches from the dataset. The model was trained using a Convolutional Neural Network (CNN) architecture (VGG 16- an object detection and classification algorithm) which was able to classify images of different categories with high accuracy. The model’s hyper- parameters such as learning rate, batch size and the number of layers were tuned with grid search and random search techniques. The performance of the model was evaluated with the application of the Sci-learn to determine the: accuracy, precision, recall and F1 score. The obtained values were compared with the existing baseline method of measuring the degree of scoliosis curve (Cobb angle) in recent times such as the use of measuring ruler. The results obtained revealed that the model achieved higher accuracy, precision, recall and F1 score of 88.23, 89.84, 86.95 and 88.49%, respectively, compared with existing baseline method of lower accuracy, precision, recall and F1 score of 80.12, 82.55, 81.55 and80.00%, respectively. The model produced predictive outputs in the form of probabilities or confidence scores for each input X-ray image with the higher scores indicated a higher probability of scoliosis, while lower scores suggested a lower likelihood. These collectively demonstrated the model's balanced performance in identifying scoliosis cases while minimizing false positives and false negatives. The model's precision ensured accurate positive predictions, while high recall ensured effective detection of actual scoliosis cases. The machine learning based model detected and classified scoliosis from X-ray images.