Efficient slice anomaly detection network for 3D brain MRI Volume.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Neuroscience

Background:

• Anomaly detection methods struggle with natural images and medical data due to ambiguous definitions of normal and abnormal.
• Current 3D brain MRI anomaly detection relies on memory-intensive, time-consuming 3D convolutional neural networks, often yielding noisy results.

Purpose of the Study:

• To develop an efficient and accurate framework for anomaly detection in 3D brain MRI volumes.
• To overcome the computational and accuracy limitations of existing 3D reconstruction-based models.

Main Methods:

• Proposed Simple Slice-based Network (SimpleSliceNet) using a pre-trained 2D feature extractor.
• Aggregated 2D slice features for 3D volume anomaly detection.
• Integrated conditional normalizing flow and contrastive loss for enhanced accuracy.

Main Results:

• SimpleSliceNet demonstrated improved performance and adaptability for brain MRI anomaly detection.
• Outperformed state-of-the-art 2D and 3D models in accuracy, memory usage, and time consumption for large-scale 3D brain volumes.

Conclusions:

• SimpleSliceNet offers a computationally efficient and highly effective solution for 3D brain MRI anomaly detection.
• The framework shows significant potential for clinical applications requiring precise identification of deviations in brain scans.