Machine learning in multi-spectral thermal imaging for enhanced detection of neurological disorders through thermoplasmonics

Thermal imaging is employed as a non-invasive diagnosing instrument in various medical practices, especially neurology. This paper delineates a systematic methodology for the development and evaluation of a multi-spectral thermal imaging system for detecting neurological diseases.

The proposed methodology incorporates three spectral types, specifically Long-Wave Infrared (LWIR), Mid-Wave Infrared (MWIR), and Near-Infrared (NIR), along with artificial neural networks (ANNs). This comprehensive integration enables several critical capabilities: First, the imaging of both surface and the deep tissues’ temperature changes, offering information about the neural activity at varying depths. Second, it enables the identification of low-contrast temperature changes detected with high-resolution. Third, it facilitates measuring the level of neural activity based on the thermal profile data.

This system demonstrates outstanding diagnostic metrics, such as an AUC of 0.923 (95 % CI: in addition, the specificity index is 0, and sensitivity ranges from 0.897 to 0.949, and the F1 score ranges from 0.891 to 0.917), rendering it clinically viable. It also achieved a moderate diagnostic accuracy of approximately 88.6 % for various neurological disorders.

This study contributes towards the further advancement of thermal imaging technology for diagnosing neurological disorders in more efficient and non-invasive manners, thereby enhancing the early screening potential for various neurological pathologies.