Article
Infrared spectroscopy for rapid analysis of native brain tumour biopsies during surgery – a study with 821 patients
Infrarotspektroskopie zur schnellen, operationsbegleitenden Analyse von nativen Hirntumorbiopsien – eine Studie mit 821 Patienten
Search Medline for
Authors
-
Ortrud Uckermann
- UKD Dresden, Neurochirurgie, Dresden, Deutschland
-
Roberta Galli
- Technische Universität Dresden, Clinical Sensing and Monitoring, Dresden, Deutschland
-
Gabriele Schackert
- UKD Dresden, Neurochirurgie, Dresden, Deutschland
-
Gerald Steiner
- Technische Universität Dresden, Clinical Sensing and Monitoring, Dresden, Deutschland
-
Matthias Kirsch
- UKD Dresden, Neurochirurgie, Dresden, Deutschland; Asklepios Kliniken Schildautal, Seesen, Deutschland
| Published: | June 4, 2021 |
|---|
Outline
Text
Objective: Optical biopsies were suggested for intraoperative brain tumor delineation and diagnosis. Particularly, infrared spectroscopy was shown to identify primary and secondary brain tumors, and is well compatible with clinical routines, as it is fast, label-free and robust. However, studies have been limited to analysis of cryosections and small sample numbers. Here, we aim to assess to potential of infrared spectroscopy for clinical exploitation by approximation of future in situ applications and investigation of the influence of interpatient variability.
Methods: We analyzed fresh biopsies of 821 patients (primary and secondary brain tumors, and non-tumor brain samples from surgery for the treatment of pharmacoresistant epilepsy). Infrared spectra were acquired within minutes after removal, reduced to the water-free range (1000-1480 cm-1) and normalized. Principal component (PC) analysis was performed on the training set consisting of spectra obtained on GBM IV (n=173) and non-tumor tissue (n=45) and the PC scores were used to develop a classification strategy based on linear discriminant analysis. For all other spectra (test set), PC scores were calculated based on the PC loadings of the training set. Subsequently, those were subjected to the classification developed on the training set (GBM versus non-tumor).
Results: The selection of principal components 1, 2, 3, 7, 10, 11, 12, 13, 14, 15 resulted in maximal correct rates for reclassification of the training set and those were therefore used for our approach. Correct rates for discrimination of samples of the test set were 84% for GBM (220 patients), 90% for brain metastases (n=177) and 84 % for a mixed group of less frequent tumor entities (n=88). Importantly, all samples of non-tumor brain tissue were correctly recognized (12 patients). Glioma WHO I-III (n=135) were not reliably recognized; possible causes are the inter-patient variability and the limited availability of non-tumor tissue.
Conclusion: Here, we show that infrared spectroscopy can help to identify tumor borders during surgery. Interestingly, a large range of different tumors was recognized using only spectra of GBM and non-tumor brain as training for the classification strategy. However, to what extent the technology is able to address infiltrative glioma remains an open question and requires in situ studies that will allow the comparison spectral signatures of non-tumor and tumor tissue of the same patient.
