Cerebral Micro-Bleeding Detection Based on Densely Connected Neural Network.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Cerebral micro-bleedings (CMBs) are small chronic brain hemorrhages with significant long-term consequences, including disability and cognitive impairment.
• Early and accurate detection of CMBs is crucial for effective treatment and management.
• Limited labeled data poses a challenge for traditional machine learning models in CMB detection.

Purpose of the Study:

• To develop an accurate and efficient method for detecting cerebral micro-bleedings (CMBs) using deep learning.
• To address the challenge of limited labeled data in CMB detection through transfer learning.
• To improve upon existing state-of-the-art methods for CMB identification.

Main Methods:

• Employed a Densely Connected Neural Network (DenseNet) as the foundational algorithm for transfer learning.
• Utilized a sliding window technique to generate subsamples from original brain images for training and testing.
• Implemented a cost matrix to address data imbalance issues during model training.

Main Results:

• Achieved a classification accuracy of 97.71% in detecting cerebral micro-bleedings.
• Demonstrated superior performance compared to existing state-of-the-art methods.
• The proposed transfer learning approach effectively handled limited labeled data.

Conclusions:

• The DenseNet-based transfer learning model offers a highly accurate solution for early detection of CMBs.
• This method shows significant potential for clinical application in diagnosing and managing conditions related to brain hemorrhages.
• The study highlights the effectiveness of advanced deep learning techniques in overcoming data limitations in medical image analysis.