Reconstruction of hyperspectral cutaneous data from an artificial neural network-based multispectral imaging system

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Biology

Background:

• Accurate in vivo skin lesion analysis is crucial for diagnosing conditions like skin cancer and inflammatory diseases.
• Current imaging techniques may lack the detailed spectral and spatial resolution needed for comprehensive assessment.
• Hyperspectral imaging offers rich data but requires advanced reconstruction methods.

Purpose of the Study:

• To develop an integrated MultiSpectral Imaging (MSI) system capable of generating hyperspectral cubes for in vivo skin lesion imaging.
• To implement a neural network-based algorithm for reconstructing hyperspectral data from multispectral images.
• To provide a tool that combines spectral and spatial information for improved skin disorder diagnosis and follow-up.

Main Methods:

• An MSI system was developed using a CCD camera, a rotating filter wheel with seven interference filters, a light source, and a computer.
• Multispectral images of skin lesions were acquired in vivo.
• A software module employing a neural network algorithm with heteroassociative memories was used to reconstruct hyperspectral cubes from the multispectral images.

Main Results:

• The developed MSI system successfully generates multispectral images.
• The reconstruction software effectively produces hyperspectral cubes from multispectral data.
• The resulting hyperspectral cubes contain detailed skin optical reflectance spectral data and bidimensional spatial information.

Conclusions:

• The integrated MSI system provides a novel approach for in vivo skin lesion analysis.
• The combination of spectral and spatial data from hyperspectral cubes is expected to improve diagnostic accuracy and patient follow-up for skin disorders.
• This technology holds promise for advancing the management of conditions ranging from skin cancer to inflammatory diseases.