Related Experiment Video
Updated: Jan 11, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Convolutional neural networks for automatic tuber segmentation and quantification of tuber burden in tuberous
Iván Sánchez Fernández1, Matheus D Soldatelli2, Gillian N Miller3
1Localization Laboratory, Division of Epilepsy and Clinical Neurophysiology, Department of Neurology, Boston Children's Hospital, Harvard Medical School, Boston, Massachusetts, USA.
A new automated algorithm using convolutional neural networks (CNNs) accurately segments brain tubers and quantifies their volume in patients with tuberous sclerosis complex (TSC). This AI tool matches expert neuroradiologist performance, improving research reproducibility.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neurology
Background:
- Tuberous Sclerosis Complex (TSC) is a genetic disorder characterized by the growth of tubers in the brain.
- Accurate tuber segmentation and volume quantification are crucial for assessing disease burden and progression in TSC.
- Current manual quantification by neuroradiologists is time-consuming and subject to inter-observer variability.
Purpose of the Study:
- To develop a fully automated algorithm for tuber segmentation and volume quantification in brain MRIs of TSC patients.
- To achieve performance comparable to a gold standard human neuroradiologist.
- To provide an objective and reproducible tool for TSC research.
Main Methods:
- A convolutional neural network (CNN) was trained and validated using brain MRI data from 196 TSC patients.
- Segmentation performance was evaluated using the Dice-Sørensen similarity coefficient (DSSC).
- Tuber burden quantification was assessed using the Spearman correlation coefficient against a neuroradiologist's manual measurements.
Main Results:
- The automated algorithm achieved a high DSSC of 0.820 for whole-brain tuber segmentation.
- Tuber volume quantification by the CNN showed a near-perfect correlation (Spearman coefficient = 0.984) with the neuroradiologist's gold standard.
- The model demonstrated strong performance across all brain lobes, with DSSC values ranging from 0.817 to 0.856.
Conclusions:
- A publicly available CNN model provides accurate and automated tuber segmentation and quantification for TSC.
- The algorithm significantly enhances objectivity and reproducibility in TSC research.
- This tool has the potential to standardize tuber burden assessment across different institutions.
