Related Experiment Video
Updated: Jun 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Real-world application of a 3D deep learning model for detecting and localizing cerebral microbleeds
So Yeon Won1,2, Jun-Ho Kim3, Changsoo Woo4
1Department of Radiology and Center for Imaging Sciences, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, Republic of Korea.
Acta Neurochirurgica
|September 26, 2024
Summary
A 3D deep learning model significantly improved the detection and localization of cerebral microbleeds (CMBs), especially for radiology residents. This AI tool shows promise for clinical use in diagnosing and planning treatments for CMBs.
Area of Science:
- Neurology
- Radiology
- Artificial Intelligence
Background:
- Cerebral microbleeds (CMBs) detection and localization are vital for disease diagnosis and treatment planning.
- Current detection methods are labor-intensive, time-consuming, and challenging due to visual mimicry.
- A 3D deep learning model was developed to address these challenges in CMB detection and localization.
Purpose of the Study:
- To validate the performance of a 3D deep learning model for detecting and anatomically locating CMBs.
- To assess the model's effectiveness in real-world clinical settings.
- To compare diagnostic performance with and without AI assistance.
Main Methods:
- A study involving 33 patients (21 with CMBs, 12 without) was conducted.
- Three readers (neuroradiologist, radiology resident, neurosurgeon) independently reviewed SWIs to detect CMBs and categorize their locations.
- Readers reassessed the datasets with AI assistance after a washout period, and performance was compared.
Main Results:
- Readers using the AI assistant generally showed higher sensitivity per lesion compared to those without AI.
- Radiology residents experienced a statistically significant increase in sensitivity when using the AI model.
- The AI model accurately categorized the anatomical locations of all CMBs, with no significant increase in false positives per patient.
Conclusions:
- The 3D deep learning model shows significant potential for detecting and anatomically localizing CMBs.
- The AI tool demonstrated improved performance, particularly for less experienced readers.
- Further research with larger, diverse populations is needed to confirm clinical utility in real-world scenarios.
