Computer-aided automated grading of brain tumors on MR spectroscopy signals

Brain tumors have been increasing rapidly in recent years as in other tumor types. In particular, a large number of patients may die if it is too late to diagnose malignant brain tumors known as cancer. Therefore, early and accurate diagnosis is vital in cancer treatment. Magnetic resonance imaging (MRI) and pathological examinations are the most common methods used in the detection of brain tumors. Pathological invasive methods like biopsy carry various risks such as disease and death. As a result of this, research and studies on non-invasive methods such as MRI and MR spectroscopy have become widespread in recent years. In this study, the classification of brain tumors at different grades is performed using LSTM (Long Short Term Memory) neural networks on Magnetic Resonance Spectroscopy (MRS) data obtained from patients with multiple brain tumors and healthy patients. As a result of this study, which proposes a method based on a computer assisted automatic diagnosis system on MR spectroscopy signals in dataset, obtained from 179 patients, it has been observed that grading is achieved with average 98.33% classification results performed between people without tumors and patients with benign and malignant tumors.