Methods and Approaches for Predicting Speech Impairment After Brain Tumor Surgery

Brain tumors are one of the most serious diseases in neurosurgery. It is usually located in the frontal and temporal lobes and affect part of the speech areas. Surgery to remove a brain tumor can cause speech impairment in patients. The article describes an algorithm for predicting speech changes brain tumor removal operations based on machine learning models. The results of electroencephalography, which were obtained before the operation, are fed to the input of the algorithm. At the output, the forecast is whether there will be a deterioration in speech or not. The data were collected and provided by N. N. Burdenko National Medical Research Center of Neurosurgery. The development of a model for predicting speech disorders consists of several stages: preliminary preparation of data, construction of a set of features, training of the model. To build a set of features based on which the model was trained, the following methods were used: feature extraction based on statistical approaches, discrete wavelet transform and signal-to-image conversion for further feature extraction using the Resnet-18 neural network. The best result was shown by an approach based on image classification. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.