JMIR Publications

ABSTRACT

Background:

Computer aided diagnosis (CAD) systems, particularly neural networks, are gaining popularity in assisting in clinical diagnoses. Various neural network models have thus far been proposed that have the potential to aid radiologists with earlier brain tumor diagnosis from magnetic resonance imaging (MRI) scans. As with many other diseases, early detection of brain tumors can pave the way for more treatment options, thus impacting the outcome of the tumor.

Objective:

We aim to explore the use of neural networks by clinicians to aid in brain tumor diagnosis from MRI scans by reviewing articles presenting novel artificial intelligence (AI)-based models, particularly convolutional neural networks (CNNs), and highlighting strengths and limitations of current methods.

Methods:

We reviewed studies of novel CNN-based brain tumor diagnosis models from the PubMed database with respect to implementation details and performance of models.

Results:

The review identified several categories of CNN models: transfer learning, conventional CNN architecture with minor variations, and hybrid architecture. All models performed remarkably well when tested on brain MRI scan datasets, either to classify the presence/absence of brain tumor, or classification of the type of brain tumor.

Conclusions:

In this paper, we present neural network-based models that were used to detect and/or classify various types of brain tumors with relative success. We provide details of the various models and highlight the strengths and limitations of current methods, demonstrating that the use of artificial intelligence in clinical decision making from imaging data holds promise and will continue to grow with advanced technological capabilities.