Related Experiment Video
Updated: Aug 3, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
Vision Transformers, Ensemble Model, and Transfer Learning Leveraging Explainable AI for Brain Tumor Detection and
IEEE Journal of Biomedical and Health Informatics
|April 12, 2023
Summary
This study introduces a novel transfer learning model (IVX16) for multiclass brain tumor classification from MRI images, achieving 96.94% accuracy. The research aims for faster, more reliable tumor detection and type identification to improve patient treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- Brain tumors, abnormal tissue growth, cause significant neurological damage.
- Magnetic Resonance Imaging (MRI) is crucial for brain tumor detection, but expert interpretation can be subjective and time-consuming.
- Accurate tumor classification is vital for timely and effective treatment initiation.
Purpose of the Study:
- To develop an automated, unbiased, and reliable method for multiclass brain tumor classification using deep learning.
- To compare the performance of various deep learning architectures for this task.
- To propose and evaluate a novel transfer learning model (IVX16) for enhanced brain tumor classification.
Main Methods:
- Investigated the performance of deep learning (DL) models: VGG16, InceptionV3, VGG19, ResNet50, InceptionResNetV2, and Xception on a dataset of 3264 MRI images.
- Developed a transfer learning (TL) based multiclass classification model named IVX16, integrating the top three performing TL models.
- Utilized Explainable AI (XAI) for model evaluation and compared results with Vision Transformer (ViT) models.
Main Results:
- Achieved peak accuracies: VGG16 (95.11%), InceptionV3 (93.88%), VGG19 (94.19%), ResNet50 (93.88%), InceptionResNetV2 (93.58%), Xception (94.5%), and the proposed IVX16 model (96.94%).
- The IVX16 model demonstrated superior performance in multiclass brain tumor classification.
- Explainable AI provided insights into model validity, and Vision Transformer models were benchmarked against the proposed solution.
Conclusions:
- Deep learning, particularly transfer learning models like IVX16, offers a promising approach for accurate and efficient multiclass brain tumor classification.
- The IVX16 model shows potential for improving diagnostic speed and reliability in neuro-oncology.
- Further research comparing advanced architectures like ViT with ensemble and TL models is warranted for clinical translation.
