Enhancing Medical X-Ray Image Classification with Neutrosophic Set Theory and Advanced Deep Learning Models

Abstract

The classification of medical images presents significant challenges due to the 
presence of noise, uncertainty, and indeterminate information. Traditional deep 
learning models often struggle to manage this, leading to reduced diagnostic 
accuracy, especially when dealing with low-quality or ambiguous conditions. This 
paper proposes a hybrid approach that integrates Neutrosophic Set (NS) theory with 
deep learning models to enhance X-ray image classification under uncertain 
conditions. NS theory introduces three domains: True (T), Indeterminate (I), and 
False (F) to manage image uncertainty and noise, allowing deep learning models to 
better interpret complex, ambiguous visual information. To evaluate the approach, 
five state-of-the-art deep learning models—MobileNet, ResNet50, VGG16, 
DenseNet121, and InceptionV3 are utilized, and their performance was evaluated 
on two different medical image datasets: Cervical spine injuries detection and chest 
disease classification. The results indicate that models trained on NS-transformed 
data, particularly DenseNet and MobileNet, yield superior outcomes compared to 
those trained on the original data, achieving significantly higher accuracy, precision, 
and recall. This demonstrates that incorporating NS theory into deep learning 
models significantly enhances their ability to classify uncertain and noisy X-ray 
images, providing a robust solution for improving diagnostic accuracy in medical 
imaging. 

DOI: 10.5281/zenodo.14880147