Article
Clinical translation of infrared spectroscopy for neurosurgery – Classification of various brain tumours based on the spectral signature of glioblastoma
Klinische Translation von Infrarotspektroskopie für die Neurochirurgie: Klassifizierung verschiedener Hirntumore auf der Grundlage der spektralen Signatur des Glioblastoms
Search Medline for
Authors
Published: | May 25, 2022 |
---|
Outline
Text
Objective: Vibrational spectroscopy including infrared and Raman spectroscopy is envisioned as innovative diagnostic tool for neurosurgery. Despite promising in situ application during glioma surgery in 2015 (Jermyn et al, Sci Transl Med 274ra19), clinical translation and routine application in neurosurgery has not been archived. Therefore, we applied infrared spectroscopy to a large set of human brain tumors to evaluate the potential of the technique for discrimination of tumor and non-tumor tissue in a clinical setting.
Methods: Samples of brain tumors and non-tumor brain tissue were obtained during routine surgery from 816 patients and probed by attenuated total refection infrared (ATR-IR) spectroscopy. IR spectra were acquired from unprocessed tissue within minutes after resection (median 6 min, 25% percentile: 3 min, 75% percentile: 10 min). They were reduced to the spectral region 1000 – 1480 cm-1 followed by a baseline correction and normalization. Spectral features of brain tumors were compared to those of non-tumor brain by analysis of band intensities and linear discriminant analysis.
Results: Tissue-specific spectral variations were identified for all types of brain tumors. IR bands related to glyco- and phospholipids (1050, 1200 cm-1) were significantly reduced in tumors while protein assigned bands (1240, 1400 cm-1) were increased. However, the marked overlap among groups hindered the exploitation of band intensities for diagnostics. Dimensionality reduction using principal components 1-14 qualified to preserve comprehensive spectral information and was used to develop a classifier for recognition of GBM. All twelve non-tumor samples and 118/134 (88%) of GBM samples were correctly recognized using this approach. Moreover, the GBM-trained classification was well suited for detection of recurrent/secondary glioma (95/115 samples), pilocytic astrocytoma (all 6 samples), other primary brain tumors (80/87 samples) and for brain metastases (163/175 samples). Half of glioma WHO II/III were misclassified.
Conclusion: We show a considerable similarity of spectroscopic signatures of different brain tumor types and propose the exploitation spectroscopic data of common tumors for recognition of other / rare tumors. The concept of an universal spectroscopic signature of brain tumors might be likewise valid for other diagnostic approaches that are based on spectroscopic tissue analysis including Raman spectroscopy and can accelerate clinical translation.