Non-contact Laser Induced Fluorescence for Intraoperative Classification of Meningioma, Glioma and Healthy Brain Tissue

Introduction: A near real time method of tumor identification for surgical staff is needed as current options such as Intraoperative MRI are expensive, time-intensive, and available only at advanced care facilities. The development of a portable, low-cost, non-contact laser induced fluorescence spectroscopy device has led to the creation of the “TumorID”. Importantly, the TumorID relies on tissue autofluorescence, therefore no additional fluorophores are used. Using an excitation wavelength of 405nm, the TumorID uses a single 0.5s scan to collect data. This study shows the TumorID’s applicability with intraoperative classification of tumor types from healthy tissue.

Methods: Tissue samples resected during the normal course of a neurosurgical procedure were scanned by the TumorID directly off the sterile field after the appropriate samples were sent to pathology. Points spaced 1mm away from each other were scanned for 0.5s with laser power of 180mW. The laser does not affect tissue histology. Over the course of the study, spectral data were collected for Meningioma (n=5 humans, 73 scans), Glioma (n=6 humans, 172 scans), and healthy brain (n=5 humans, 70 scans) tissue. The data were normalized with respect to the largest intensity value greater than 425nm in each scan and then smoothed to decrease random noise in the data. These data were then split into 70% training, 30% testing data sets to train a logistic regression model.

Results: The model had an F1-Score, which weighs both the precision and recall of the model, of 91% for healthy tissue, 92% for Glioma, and 86% for Meningioma. The model had an overall accuracy of 90.5%.

Conclusion : This study shows the capabilities of using laser induced non-contact endogenous fluorescence spectroscopy and logistic regression as a tissue classification model to not only differentiate healthy tissue from tumor tissue but also to find differences between tumor types.