According to a study published in Radiology: Artificial Intelligence. Additionally, this was achieved from a single scan per patient without the need for any additional manual input.
In this retrospective study, Satrajit Chakrabarty, a Ph.D. candidate in electrical engineering at Washington University in St. Louis, and his colleagues developed a three-dimensional (3D) convolutional neural network model to classify MRI scans into a healthy class and six tumor classes: high-grade glioma, low-grade glioma, brain metastases, meningioma, pituitary adenoma and acoustic neuroma.
“This is the first study to treat the most common types of intracranial tumors and directly determine tumor class as well as detect tumor absence from a 3D MRI volume,” said the authors. “Without the help of a tumor or a manually segmented bounding box, the convolutional neural network model… could classify six types of brain tumors and discriminate
healthy from pathological analyzes from a single post-contrast T1-weighted analysis per patient.
Researchers included 2,105 T1-weighted post-contrast preoperative MRIs from four publicly available datasets, and 1396 scans were used to train the model, while 361 scans were assigned to an internal test dataset. and 348 scans were assigned to an external test dataset.
On the internal test dataset, in the seven different classes, the model achieved a sensitivity of 87% to 100%, a positive predictive value (PPV) of 85% to 100%, an area under the characteristic curve of receiver operation (AUC) from 0.98 to 1.00 and precision recall curve (AUPRC) from 0.91 to 1.00. On the external test dataset, including one high grade glioma and one low grade glioma, the model achieved sensitivity of 91% to 97%, PPV of 73% to 99%, AUC of 0, 97 to 0.98 and an AUPRC of 0.9 to 1.0.
Of the seven classes discussed, more errors were observed for high-grade gliomas, low-grade gliomas and healthy gliomas. The authors explained that the reason for the misclassification between low-grade gliomas and healthy classes may be due to less improvement in the contrast of low-grade gliomas in T1-weighted post-contrast MRIs due to less disturbance of the blood-brain barrier.
“The model can be extended to other types of brain tumors or to neurological disorders that present
abnormal intensity profiles in MRI scans, ”the authors said. “The network is the first step towards the development of an artificial intelligence enhanced radiology workflow that can support image interpretation by providing quantitative information and statistics to the clinician to help improve diagnosis and the prognosis.