Researchers Develop AI Model That Classifies Brain Tumours With Single MRI Scan

According to a story published in science journal: Radiology: Artificial Intelligence , Washington University School of Medicine researchers, using Artificial intelligence has developed a model that would help classify the tumour of the brain with a single 3D MRI scan.

A new breakthrough in medicine you say?

You may be right after all!

According to a joint statement by Satrajit Chakrabarty, M.S., a doctoral student, and Daniel Marcus, Ph.D. from the  Washington University School of Medicine in St. Louis, Missouri, who formed part of the research team: “This is the first study to address the most common intracranial tumours and to directly determine the tumour class or the absence of tumour from a 3D MRI volume”.

High-grade glioma, low-grade glioma , brain metastases, meningioma, pituitary adenoma and acoustic neuroma, six of the most common types of tumour were documented through the process called his pathology, wherein tissues are surgically removed from a suspected cancer cell and examined with the use of a microscope.

Chakrabarty while analysing how MRI data would be used to automatically detect and classify brain tumours said:

“Non-invasive MRI may be used as a complement, or in some cases, as an alternative to histopathologic examination,”

The researchers from the Mallinckrodt Institute of Radiology , together with Chakrabarty built a massive, multi-institutional intracranial MRI scans 3D datasets through a convolutional neural network, obtaining pre-operative, post contrast T1-weighted MRI scans using the Brain Tumour Image Segmentation, The Cancer Genome Atlas Glioblastoma Multiforme, and The Cancer Genome Atlas Low Grade Glioma processes.

2,105 total scans were then grouped into three data subsets, with 1.396 of it for training, 261 scans for internal testing while 348 scans was for external testing. The performances of the model suing data from the external and internal MRI scans were evaluated after the first MRI scan set were used to train the convolutional neural network to separate healthy scans and unhealthy ones that has tumours, helping the classification of tumour type.

With the use of the internal testing data, a 93.35 percent accuracy was gotten from a healthy class and six tumour class with a accuracy figure of 337 out of 361.

Positive Predictive Value, with sensitivities ranging from 91 percent to 100 percent, has the probability of patients who had hitherto had a positive screening test having the disease ranging from 85 percent to 100 percent.

On the other hand, Negative Predictive Value ranged from 98 percent to 100 percent across all the classes with network attention overlapping with the tumour areas for all tumour types.

For the external test dataset, which included only two tumour types (high-grade glioma and low-grade glioma), the model had an accuracy of 91.95%.

According to Chakrabarty ,“These results suggest that deep learning is a promising approach for automated classification and evaluation of brain tumours. The model achieved high accuracy on a heterogeneous dataset and showed excellent generalization capabilities on unseen testing data.”

He further added that  the 3D deep learning model aligns more to the goal of an end-to-end, automated workflow where existing 2D models are improved upon, a process that requires radiologists to manually delineate, or characterize, the tumour area on an MRI scan before machine processing.

According to Dr. Sotiras, who is a co-developer of the model, the method can be extended to other brain tumour types or neurological disorders, potentially providing a pathway to augment much of the neuroradiology workflow.

“This network is the first step toward developing an artificial intelligence-augmented radiology workflow that can support image interpretation by providing quantitative information and statistics,” Chakrabarty added.

Reference: “MRI-based Identification and Classification of Major Intracranial Tumour Types Using a 3D Convolutional Neural Network: A Retrospective Multi-Institutional Analysis” 11 August 2021, Radiology: Artificial Intelligence.