P10.12.B DEEP LEARNING-BASED NORMALIZATION AND UNMIXING OF HYPERSPECTRAL IMAGES FOR BRAIN TUMOR SURGERY

Abstract BACKGROUND Hyperspectral Imaging (HSI) for fluorescence-guided brain tumor resection enables better visualization of differences between tissues. This can maximize brain tumor resection, improving patient outcomes. However, much of the processing in HSI uses simplified models that are unable to capture the non-linear, wavelength-dependent phenomena that must be modeled for accurate recovery of fluorophore abundances. To overcome these challenges, measured spectra are normalized to account for artifacts and spectrally unmixed to isolate the protoporphyrin IX (PpIX) signal, which indicates malignant tissue. Existing methods are simplistic and based on phantoms and are unable to account for nonlinear effects such as multiple scattering or the inhomogeneous optical properties of the tissue. We propose a deep learning (DL)-based pipeline encompassing normalization and unmixing, which can fully account for the nonlinear effects and produce more accurate distributions of fluorophore abundances. MATERIAL AND METHODS This study proposes two deep learning models for correction and unmixing, which can capture these effects. Both models use autoencoder-like architectures. While one is trained with protoporphyrin IX (PpIX) concentration labels, the other is semi-supervised, learning hyperspectral unmixing self-supervised and then learning to correct the spectra for optical and geometric effects using a white-light reflectance spectrum in a few-shot manner. The models were evaluated on phantom and pig-brain data with known PpIX concentration. RESULTS The supervised and semi-supervised models achieved Pearson correlation coefficients (phantom, pig-brain) between known and computed PpIX concentrations of (0.997, 0.990) and (0.98, 0.91), respectively, whereas the classical approach achieved (0.93, 0.82). On human data, the semi-supervised model gives qualitatively more realistic results than the classical method, better reducing the variance in PpIX abundance over biopsies that should be relatively homogeneous. CONCLUSION These results show promise for using deep learning to improve HSI in fluorescence-guided neurosurgery.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.