Raman difference spectroscopy and U-Net convolutional neural network for molecular analysis of cutaneous neurofibroma

Area of Science:

• Biomedical Optics
• Artificial Intelligence in Medicine
• Oncology

Background:

• Cutaneous neurofibromas in Neurofibromatosis type 1 (NF1) cause significant aesthetic, painful, and functional issues.
• Current surgical removal relies on subjective margin assessment, leading to incomplete resection and recurrence.
• Objective, real-time diagnostic tools are needed to guide surgical ablation and improve patient care.

Purpose of the Study:

• To introduce Shifted-Excitation Raman Spectroscopy (SERS) combined with deep neural networks (DNNs) for objective, real-time diagnosis of cutaneous neurofibromas.
• To evaluate the efficacy of this technique in discriminating between neurofibromas and adjacent healthy tissues.
• To provide a foundation for guided surgical ablation in NF1 patients.

Main Methods:

• Ex vivo analysis of 34 surgical pathological specimens (9 cutaneous neurofibromas, 25 physiological tissues).
• Application of Shifted-Excitation Raman Spectroscopy (SERS) for spectral data acquisition.
• Deep neural network (U-Net) for spectral processing, combined with Principal Component and Linear Discriminant Analysis for classification.

Main Results:

• High diagnostic performance in discriminating cutaneous neurofibromas from physiological tissues.
• Achieved sensitivity of 100%, specificity of 97.3%, and overall classification accuracy of 97.6%.
• Spectral peaks successfully assigned to their respective molecular origins, validating the technique.

Conclusions:

• Shifted-Excitation Raman Spectroscopy (SERS) with artificial intelligence offers a promising non-invasive approach for diagnosing cutaneous neurofibromas.
• This optical technique has the potential to significantly improve both the diagnosis and surgical treatment of NF1.
• Objective, real-time feedback can guide surgeons, minimizing residual tumor and recurrence rates.