Deep learning-based automatic delineation of the hippocampus by MRI: geometric and dosimetric evaluation

Area of Science:

• Radiotherapy and Medical Imaging
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Whole brain radiotherapy (WBRT) can lead to cognitive impairments.
• Hippocampal avoidance during WBRT is a strategy to mitigate cognitive side effects.
• Manual delineation of the hippocampus for radiotherapy planning is labor-intensive and challenging.

Purpose of the Study:

• To develop and validate an automatic hippocampal delineation (AD) method using convolutional neural networks.
• To assess the accuracy and clinical feasibility of the AI-driven AD tool compared to manual delineation (MD).

Main Methods:

• A 3D convolutional neural network was trained on 175 manually delineated hippocampus MRI datasets.
• The AD tool was tested on three independent cohorts with varying MRI slice thicknesses.
• Virtual radiation plans were generated using both AD and MD hippocampi to evaluate clinical applicability.

Main Results:

• The AD tool achieved high accuracy, with Dice similarity coefficients (DSC) ranging from 0.76 to 0.86 and Average Hausdorff Distances (AVD) from 0.18 to 0.31 cm across cohorts.
• Performance was superior in the cohort with 1-mm slice thickness 3D T1 MRI.
• No significant differences were found between radiotherapy plans generated with AD and MD hippocampi.

Conclusions:

• Deep learning-based automatic hippocampal delineation is accurate and effective.
• This AI approach has the potential to improve delineation precision and reduce the manual workload in radiotherapy planning.
• The AD method shows clinical feasibility for preserving cognitive function during WBRT.