Related Experiment Video
Updated: Mar 11, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.7K
Classification of CT brain images based on deep learning networks
Xiaohong W Gao1, Rui Hui2, Zengmin Tian2
1Department of Computer Science, Middlesex University, London NW4 4BT, UK.
Computer Methods and Programs in Biomedicine
|November 26, 2016
Summary
This study uses deep learning and convolutional neural networks (CNNs) to classify brain CT scans for Alzheimer's disease (AD) diagnosis. The advanced CNN model achieved high accuracy, outperforming other methods for early AD detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Computerized Tomography (CT) is a prevalent, inexpensive, and non-invasive imaging tool for brain studies.
- Despite its potential, CT has not been widely adopted for clinical Alzheimer's disease (AD) diagnosis.
- Deep learning techniques offer promising avenues for enhancing diagnostic capabilities using medical imaging.
Purpose of the Study:
- To explore the application of deep learning, specifically Convolutional Neural Networks (CNNs), for classifying brain CT images.
- To develop an advanced CNN architecture for the early diagnosis of Alzheimer's disease.
- To provide supplementary diagnostic information for AD using CT scans.
Main Methods:
- A dataset of 285 CT brain images was classified into three categories: Alzheimer's disease (AD), lesions (e.g., tumors), and normal aging.
- An advanced CNN architecture was developed, integrating both 2D and 3D CNNs to leverage the volumetric nature of CT data (slice thickness ~3-5mm).
- Fusion of 2D and 3D CNN outputs was achieved by averaging Softmax scores to consolidate information from axial slices and 3D blocks.
Main Results:
- The elaborated CNN architecture achieved classification accuracy rates of 85.2% for AD, 80% for lesions, and 95.3% for normal aging, with an overall average of 87.6%.
- The proposed hybrid 2D/3D CNN model demonstrated superior performance compared to a 2D CNN-only approach and several state-of-the-art handcrafted feature methods.
- Comparative analysis showed the hybrid CNN outperformed 2D CNN (86.3%), 2D SIFT (85.6% ± 1.10), 2D KAZE (86.3% ± 1.04), 3D SIFT (85.2% ± 1.60), and 3D KAZE (83.1% ± 0.35).
Conclusions:
- The study introduces a novel 3D deep learning approach for extracting information from both 2D slices and 3D blocks of CT images.
- The developed hybrid CNN architecture shows significant potential for improving the accuracy of Alzheimer's disease diagnosis using CT scans.
- This research highlights the impact of advanced deep learning techniques in enhancing the diagnostic utility of conventional imaging modalities like CT for neurodegenerative diseases.
