According to a recent study, a new deep learning model can help classify a brain tumor as one of six most common types using a single 3D MRI scan.
Washington: As part of a new study, researchers have developed a deep learning model that is able to classify a brain tumor as one of the six most common types based on a single 3D MRI image. The study by researchers from the Washington University School of Medicine has been published in Radiology: Artificial Intelligence.
“This is the first study to look at the most common intracranial tumors and determine tumor class or the absence of a tumor directly from a 3D MRI volume,” said Satrajit Chakrabarty, M., PhD student led by Aristeidis Sotiras, PhD and Daniel Marcus, PhD, in the Mallinckrodt Radiology Institute’s Computational Imaging Laboratory at Washington University School of Medicine in St. Louis, Missouri.
The six most common types of intracranial tumors are high grade glioma, low grade glioma, brain metastasis, meningioma, pituitary adenoma, and acoustic neuroma; each has been documented by histopathology, which requires surgical removal of tissue from the suspected cancer site and microscopic examination. For Chakrabarty, deep and machine learning approaches that use MRI data could potentially automate the detection and classification of brain tumors.
“Non-invasive MRI can be used as an adjunct or in some cases as an alternative to histopathological examination,” he said. To build their machine learning model, known as a convolutional neural network, Chakrabarty and researchers at the Mallinckrodt Institute of Radiology developed a large, multi-agency 3D intracranial MRI dataset from four publicly available sources.
In addition to the institution’s internal data, the team received preoperative and post-contrasting T1-weighted MRI scans of image segmentation of brain tumors, glioblastoma multiforme from the Cancer Genome Atlas, and low-grade gliomas from the Atlas of the Cancer. Partial data sets: 1,396 for training, 361 for internal tests and 348 for external tests. The first set of MRIs were used to train the convolutional neural network to differentiate between healthy scans and scans with tumors, and to classify tumors by type. Researchers evaluated the model’s performance using data from internal and external MRIs.
Using internal test data, the model achieved an accuracy of 93.35 percent (337 out of 361) for seven image classes (one healthy class and six tumor classes). The sensitivities were between 91 and 100 percent, the positive predictive value or the probability that patients with a positive screening test actually have the disease was between 85 and 100 percent. Negative predictive values or the probability that patients with a negative screening test actually do not suffer from the disease ranges from 98 to 100 percent in all classes. Network attention overlapped with tumor areas in all tumor types. For the external test dataset, which included only two tumor types (high grade glioma and low grade glioma), the model was 91.95% accurate.
“These results suggest that deep learning is a promising approach for automated classification and evaluation of brain tumours,” Chakrabarty said. “The model achieved high accuracy on a heterogeneous dataset and showed excellent generalization capabilities on unseen testing data.”
Chakrabarty said the 3D deep learning model comes closer to the goal of an end-to-end, automated workflow by improving upon existing 2D approaches, which require radiologists to manually delineate, or characterize, the tumour area on an MRI scan before machine processing. The convolutional neural network eliminates the tedious and labour-intensive step of tumour segmentation before classification.
Dr. Sotiras, a co-developer of the model, said it can be extended to other types of brain tumors or neurological diseases and potentially provides a way to improve the neuroradiological workflow.
“Artificial Intelligence advanced radiology workflow that can aid image interpretation by providing quantitative information and statistics “Added Chakrabarty.
