Detection of pseudo brain tumors via stacked LSTM neural networks using MR spectroscopy signals

Abstract

Magnetic resonance spectroscopy (MRS) is one of the non-invasive tools used in the detection of brain tumors. MRS provides a metabolic profile about the brain. In this profile, MRS patterns of the tumors and pseudo tumors can be similar to each other. For this reason, accurate diagnosis and classification of brain tumor is of vital importance for the patient's treatment planning. It has been widely preferred by physicians in recent years because it does not pose the risk of infection and death due to surgery like biopsy. In this study, binary classification of brain tumors and normal brain tissue with pseudo-brain tumors is achieved via deep neural networks using MRS data. For the classification of MRS signals, a stacked model based on Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (Bi-LSTM) deep neural networks is proposed. In the experimental studies in the study, MRS signals from normal brain tissue, brain tumor and pseudo-brain tumors in the INTERPRET database are used. Since the MRS data belonging to a large number of tumors and pseudo-tumors are required for training and testing of the LSTM neural networks, the number of data for the MRS dataset is increased by data augmentation methods. Training and testing of the LSTM neural networks used are performed with a repeated 5-fold cross validation and 10 repetitions for each model. As a result of this study, proposed a stacked model for computer-aided binary classification of MRS data, classification results of 93.44%, 85.56%, 88.33% and 99.23% are obtained for the classification of pseudo brain tumor with glioblastoma, diffuse astrocytoma, metastatic brain tumors and normal brain tissue, respectively. Therefore, it is confirmed that the proposed LSTM-based stacked method is successful in detecting pseudo brain tumors using MRS signals. ? 2020 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.