Supervised Machine Learning Methods and Hyperspectral Imaging Techniques Jointly Applied for Brain Cancer Classification

Abstract

Hyperspectral imaging techniques (HSI) do not require contact with patients and are non-ionizing as well as non-invasive. As a consequence, they have been extensively applied in the medical field. HSI is being combined with machine learning (ML) processes to obtain models to assist in diagnosis. In particular, the combination of these techniques has proven to be a reliable aid in the differentiation of healthy and tumor tissue during brain tumor surgery. ML algorithms such as support vector machine (SVM), random forest (RF) and convolutional neural networks (CNN) are used to make predictions and provide in-vivo visualizations that may assist neurosurgeons in being more precise, hence reducing damages to healthy tissue. In this work, thirteen in-vivo hyperspectral images from twelve different patients with high-grade gliomas (grade III and IV) have been selected to train SVM, RF and CNN classifiers. Five different classes have been defined during the experiments: healthy tissue, tumor, venous blood vessel, arterial blood vessel and dura mater. Overall accuracy (OACC) results vary from 60% to 95% depending on the training conditions. Finally, as far as the contribution of each band to the OACC is concerned, the results obtained in this work are 3.81 times greater than those reported in the literature.

Keywords: brain; classification; convolutional neural network; hyperspectral imaging; machine learning; neurosurgery; random forest; support vector machine; tumor.

Figure 1

Diagram of the designed methodology for both experiments: (a) Experiment A (top), (b) Experiment B (bottom).

Figure 2

Dataset: representation of mean spectral signatures of all 5 classes.

Figure 3

Dataset: representation of mean spectral signatures with standard deviation for each class.

Figure 4

Synthetic RGB and GT maps of patients studied.

Figure 5

CNN implemented with the filter sizes.

Figure 6

Double cross-validation procedure to train models. For simplicity, the figure only shows external and internal loops for K = 1 combination. K_n combinations follow the same procedure even though the figure only shows it for K_n = 1.

Figure 7

Diagram for both experiments: (a) Experiment A (top) and (b) Experiment B (bottom).

Figure 8

Error in % for SVM (blue), RF (red) and CNN (brown) after classifying each patient image in Experiment A.

Figure 9

Mean $ACC$ for all patient predictions and every tissue for SVM, RF and CNN algorithms obtained in Experiment A.

Figure 10

SVM, RF, CNN accuracy in tumor class for each patient in Experiment A.

Figure 11

Mean $SEN$ for all patient predictions and every tissue for SVM, RF and CNN algorithms obtained in Experiment A.

Figure 12

SVM, RF, CNN sensitivity in tumor class for each patient in Experiment A.

Figure 13

RF sensitivity for each patient and tissue in Experiment A.

Figure 14

Mean SPEC for all patient predictions and every tissue for SVM, RF and CNN algorithms obtained in Experiment A.

Figure 15

Synthetic RGBs, ground truths and classification maps of patients ID33 and ID51 obtained in Experiment A after being classified by RF, SVM and CNN.

Figure 16

Mean $ACC$ for all patient predictions and every tissue for SVM, RF and CNN algorithms obtained in Experiment B.

Figure 17

SVM, RF, CNN accuracy in tumor class for each patient in Experiment B.

Figure 18

Mean $SEN$ for all patient predictions and every tissue for SVM, RF and CNN algorithms obtained in Experiment B.

Figure 19

SVM sensibility for each patient and tissue in Experiment B.

Figure 20

Synthetic RGBs, ground truths and classification maps of patients ID33 and ID51 obtained in Experiment B after being classified by RF, SVM and CNN.