Machine Learning-Based Fluorescence Assessment for Augmented Imaging and Decision Support in Glioblastoma Resections.

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Imaging

Background:

• Glioblastoma is an aggressive brain tumor with challenging surgical resection.
• Current fluorescence-guided surgery (5-aminolevulinic acid) has limitations in sensitivity and reliability.
• Pixel-level image analysis using machine learning can potentially improve tumor visualization.

Purpose of the Study:

• To develop and evaluate a machine learning framework for enhanced fluorescence detection in glioblastoma surgery.
• To improve the sensitivity and objectivity of fluorescence-guided surgery.
• To aid in achieving maximal tumor resection while minimizing neurological deficits.

Main Methods:

• Trained machine learning classifiers (SVM, Naïve Bayes, Neural Network) and Variational Autoencoders (VAEs) on synthetic and real intraoperative microscope images.
• Utilized pixel-level analysis to detect Protoporphyrin IX (PPIX) fluorescence.
• Compared the model's detection sensitivity against expert surgeon annotations.

Main Results:

• The machine learning model demonstrated high detection rates with excellent specificity (>99.9%) on real surgical data.
• The model identified larger fluorescent tumor areas compared to surgeon annotations.
• A regression model accurately quantified PPIX concentration in synthetic samples (R2=0.92).

Conclusions:

• The proposed AI framework significantly enhances fluorescence detection sensitivity and objectivity in glioblastoma surgery.
• This technology can improve surgical decision-making, leading to safer and more complete tumor resections.
• The approach holds promise for advancing image-guided neurosurgical techniques.