Predicting Meningioma Recurrence Using Spectrochemical Analysis of Tissues and Subsequent Predictive Computational Algorithms

Lilo, Taha Luay (2022) Predicting Meningioma Recurrence Using Spectrochemical Analysis of Tissues and Subsequent Predictive Computational Algorithms. Doctoral thesis, University of Central Lancashire.

[thumbnail of Thesis]
Preview
PDF (Thesis) - Submitted Version
Available under License Creative Commons Attribution Non-commercial.

8MB

Digital ID: http://doi.org/10.17030/uclan.thesis.00047977

Abstract

Meningiomas are the most common types of tumour of the central nervous system (CNS) and are classified as WHO grades (1,2 ,&3) depending on histological sub-type, tumour growth rate and the likelihood of recurrence. The majority of meningioma are benign, yet, around 10% will recur following resection. Variation in follow-up of patients comes with significant clinical, logistical, and financial implications, hence, the search for predictors for meningioma recurrence has become an increasingly urgent research topic.
AIM The aim was to assess the suitability of biospectroscopy sensor-based techniques Fourier-transform infrared (FTIR) and Raman spectroscopy for analysis of meningioma tissues to accurately segregate patients with benign meningioma (WHO 1 &2) into either high-risk group for recurrence or low risk group based on the spectrochemical signature.
METHODS Patients with convexity meningioma (n=99), Simpson grade 1 or 2 only and WHO grade 1 (n=70) or grade 2 (n=24) with a minimum 5 years follow up (n=5 recurrence) were consented for study. Formalin-fixed paraffin-embedded (FFPE) were sectioned and de-waxed prior to ATR-FTIR or Raman spectrochemical analyses. Derived spectral datasets were then explored for discriminating features via multivariate analysis and machine learning algorithms, such as principal component analysis linear discriminant analysis (PCA-LDA) and partial least squares discriminant analysis (PLS-DA). Three-dimensional (3D) discriminant analysis techniques were also used to analyse Raman hyperspectral tissue images in a (3D) fashion.
RESULTS: WHO grade 1 verses grade 2 meningioma samples and those that recurred from those that did not recur were accurately and blindly segregated. For the ATR-FTIR data, PLS-DA gave the best results where grade 1 and grade 2 meningiomas were discriminated with 79% accuracy, 80% sensitivity and 73% specificity; while grade 1 vs. grade 1 recurrence and grade 2 vs. grade 1 recurrence were discriminated with 94% accuracy (94% sensitivity and specificity) and 97% accuracy (97% sensitivity and 100% specificity), respectively. For the Raman data, the classical spectral analysis after extracting each spectrum from the Raman imaging area achieved best classification performances by using principal component analysis-quadratic discriminant analysis (PCA-QDA) and successive projections algorithm quadratic discriminant analysis (SPA-QDA), resulting in accuracies of 96.2%, sensitivities of 85.7% and specificities of 100% using both algorithms. For the Raman 3D image data, 3D principal component analysis quadratic discriminant analysis (3D-PCA-QDA) was able to distinguish grade 1 and grade 2 meningioma samples with 96% test accuracy (100% sensitivity and 95% specificity), and most recurrence samples were predicted as grade 2 which have higher likelihood of recurrence.
DISCUSSION Several wavenumbers were identified as possible biomarkers towards tumour differentiation, associated with lipids, protein, DNA/RNA, and carbohydrate alterations. For Raman spectroscopy, the following wavenumbers were found to be associated with class differentiation: 850 cm-1 (amino acids or polysaccharides), 1130 cm-1 (phospholipid structural changes), the region between 1230–1360 cm-1 (Amide III and CH2 deformation), 1450 cm-1 (CH2 bending), and 1858 cm-1 (C=O stretching). These findings highlight the potential of Raman microspectroscopy imaging for determination of meningioma tumour grades.
CONCLUSION Reagent-free, non-destructive, and low-cost ATR-FTIR and Raman spectroscopy techniques could give predictive information towards meningioma grade discrimination and the propensity of meningioma to recur. This has enormous clinical potential with regards to being developed for intra-operative real-time assessment of disease. In addition, by building a predictive reoccurrence model in advance, it would be possible to predict the best treatment for the patient according to the likelihood of tumour reoccurrence.


Repository Staff Only: item control page