Pediatric Brain Tissue Segmentation Using a Snapshot Hyperspectral Imaging (sHSI) Camera and Machine Learning...

Area of Science:

• Medical imaging
• Oncology
• Surgical technology

Background:

• Pediatric brain tumors are a significant cause of childhood cancer.
• Accurate intraoperative assessment of tumor margins is crucial but challenging.
• Current biopsy methods for margin assessment can be time-consuming and lack precision.

Purpose of the Study:

• To evaluate the efficacy of snapshot hyperspectral imaging (sHSI) for real-time segmentation of healthy brain tissue from lesions in pediatric patients undergoing tumor resection.
• To assess the performance of different sHSI spectral bands (RGB, visible, infrared) in differentiating tumor margins.

Main Methods:

• sHSI data (RGB, visible, infrared) were collected from four pediatric patients during brain tumor resection surgery using a system integrated into an operating microscope.
• A random forest classifier was employed for data analysis and tissue segmentation.
• Performance was evaluated using intersection of union (IoU) for segmentation and specificity for tumor detection.

Main Results:

• The RGB sHSI model achieved the highest average IoU (0.76).
• Tumor segmentation demonstrated high specificity, with the RGB model reaching 0.996, followed by infrared (0.93) and visible (0.91) models.
• Despite a small dataset, sHSI successfully segmented healthy tissues from lesions with high accuracy.

Conclusions:

• Snapshot hyperspectral imaging is a promising technology for real-time intraoperative guidance in pediatric brain tumor surgery.
• sHSI enables accurate differentiation between healthy brain tissue and tumors, potentially improving surgical outcomes.
• Further research with larger datasets is warranted to fully establish sHSI's role in neuro-oncology.