Rapid screening for autoimmune diseases using Fourier transform infrared spectroscopy and deep learning algorithms.

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Computational Biology

Background:

• Ankylosing spondylitis (AS), rheumatoid arthritis (RA), and osteoarthritis (OA) are rheumatic immune diseases that can lead to severe joint destruction and disability if untreated.
• Early diagnosis and treatment are critical for improving patient outcomes and prognosis in rheumatic immune diseases.
• Current diagnostic methods may lack the speed and accuracy required for timely intervention.

Purpose of the Study:

• To develop a rapid, non-invasive, and accurate method for differentiating between AS, RA, OA, and healthy individuals.
• To evaluate the efficacy of Fourier transform infrared spectroscopy (FTIR) combined with deep learning models for disease diagnosis.
• To identify the optimal deep learning model for distinguishing these rheumatic conditions based on serum spectral data.

Main Methods:

• Serum samples from 320 individuals (80 each of AS, RA, OA, and healthy controls) were analyzed using Fourier transform infrared spectroscopy (FTIR) in the 700-4000 cm-1 range.
• Four deep learning models (AlexNet, ResNet, MSCNN, MSResNet) were developed using machine learning algorithms to classify the FTIR spectral data.
• The MSResNet model, incorporating multi-scale convolutional modules and residual blocks, was specifically designed to enhance feature extraction and reduce spectral noise.

Main Results:

• Serum FTIR spectroscopy revealed characteristic spectral peaks related to protein and lipid components.
• The MSResNet deep learning model demonstrated superior performance compared to AlexNet, ResNet, and MSCNN.
• The MSResNet model achieved the highest diagnostic accuracy of 0.87 in differentiating between the four groups.

Conclusions:

• Serum FTIR spectroscopy combined with deep learning algorithms provides a feasible and effective auxiliary diagnostic method for AS, RA, and OA.
• This approach enables non-invasive, rapid, and accurate differentiation of these rheumatic immune diseases.
• The findings highlight the potential of advanced computational methods in improving the early diagnosis and management of rheumatic conditions.