Artificial Intelligence (AI) has enabled in pushing the boundaries of medical research. Proving that AI can help in better diagnosis and treatment of brain tumors, scientists have developed a new machine-learning algorithm that can classify tumors based on the severity.
Developed by scientists from Kyoto University's Institute for Integrated Cell-Material Sciences (iCeMS) and Indian Institute of Technology (IIT), Roorkee, the machine learning approach can classify a common form of brain tumor known as gliomas, into low and high grade, with nearly 98 percent accuracy.
"We hope AI helps develop a semi-automatic or automatic machine predictive software model that can help doctors, radiologists, and other medical practitioners tailor the best approaches for their individual patients," said Ganesh Pandian Namasivayam, a bioengineer at iCeMS, and lead author of the study.
Relying On AI For Better Diagnosis
Gliomas are a common kind of brain tumor that affects the glial cells. These are non-neural cells that provide insulation and support to neurons. Treatment for patients varies based on the aggressiveness of their tumors.
Most of the time, large quantities of data obtained from MRI scans of the tumorous growth, are used to construct a 3D image of the scanned areas by radiologists. However, several minute and useful details that can help in the effective treatment of the ailment cannot be observed with the naked eye. These include texture or the intensity of the image, and shape of the tumor.
AI algorithms come into play in the extraction of such crucial data. Using an approach known as radiomics, oncologists have been able to provide a better diagnosis for patients with gliomas. However, accuracy has still been an area of concern.
Aiming For Accuracy
This is where the new algorithm comes into play. In order to develop it, Namasivayam, a bioengineer at iCeMS, joined forces with Balasubramanian Raman, an Indian data scientist from IIT Roorkee. The machine learning approach is based on the classification of gilomas into two categories of severity based on their MRI scans –Low and High grade.
Low grade gliomas consist of slow-growing and less aggressive form of glioma tumors known as grade I pilocytic astrocytoma and grade II low-grade glioma. High grade gliomas include tumors of higher malignancy and aggressiveness known as grade III malignant glioma and grade IV glioblastoma multiforme. Therefore, the treatment of patients relies heavily on the grading of their gliomas.
Developing A New Approach
The team developed an approach called computational decision support system for glioma classification using hybrid radiomics and stationary wavelet-based features (CGHF). For the study, they used data sets comprising of MRI scans from 210 people suffering from high grade gliomas and 75 from patients with low grade gliomas.
The scientists then employed particular algorithms to extract characteristics from some of the MRI scans. In order to process the data gathered from the scans and classify the gliomas, they trained another algorithm of predictive nature. Following this, the team tested the model with the remainder of the MRI scans to analyze its accuracy. They found that the algorithm could classify the scans with 97.54 percent accuracy.
"Our method outperformed other state-of-the-art approaches for predicting glioma grades from brain MRI scans. This is quite considerable," said Namasivayam.
