Hyperspectral imaging for dermatology

A Novel Nonparametric Technique for Segmenting Multimode Hyperspectral ImagesObtained from Non-Melanoma Skin Cancer Lesions

Background

Non-melanoma skin cancer (NMSC) is the most common form of cancer. In the US over 4 million cases are diagnosed each year, resulting in more than 1.5 surgical procedures, though many are simple mole removals. Lesions or moles evolve over time. Early detection and targeted removal are crucial.

This paper was presented at the 2020 Design of Medical Devices Conference in Minneapolis, Minnesota, USA. The research aimed to improve diagnostic efficiency and efficacy with the use of hyperspectral imaging. Sensitivity and specificity are crucial in surgery. Clinicians want to avoid failing to remove cancerous tissue. Hyperspectral imaging is non-invasive and non-ionising. The optical sensors can generate precise and rapid map of the margins of lesions. That helps in determining the actual borderlines.‍

‍

Methods

The researchers conducting the study used a hyperspectral imaging system to obtain patient lesion images at different wavelengths. Different materials absorb and reflect light in unique ways. The more spectral bands the higher the signal-to-noise ratio, the more sensitive and specific is the instrument.

The research showed about 90% of NMSC can be identified during the imaging process which includes motion artifacts removal, image enhancement and topological retraction. However, differentiating between healthy and cancerous tissue remains an opportunity for further innovation. Enhancements to the signal-to-noise ratio could help better distinguish the white matter from the gray matter.‍

‍

Implications

The study indicates that the presented method of hyperspectral imaging segmentation offers high accuracy and therefore clinical potential. The research demonstrated the benefits of applying hyperspectral to detect non-melanoma, also indicating the speed and precision of recognition. A more sensitive and powerful sensor could push the sensitivity and specificity higher, providing roadmap for the clinical adoption of hyperspectral for disease detection and other biomedical applications.