Fully automated clinical target volume segmentation for glioblastoma radiotherapy using a deep convolutional neural network

Pol J Radiol, 2023 · DOI: https://doi.org/10.5114/pjr.2023.124434 · Published: January 19, 2023

Simple Explanation

The study focuses on using a special type of computer program, a deep convolutional neural network (CNN), to automatically identify the clinical target volume (CTV) in glioblastoma patients for radiotherapy planning. The CNN model was initially trained on a large dataset of glioblastoma patients to segment the gross tumor volume (GTV). This pre-trained model was then fine-tuned using a smaller, independent set of patient data to specifically identify the CTV, using both CT and MRI scans as input.

Study Duration

Not specified

Participants

259 glioblastoma patients (BraTS 2019), 37 for fine-tuning, 15 for testing (clinical dataset)

Evidence Level

Not specified

Key Findings

1
The proposed model achieved high segmentation accuracy with a Dice Similarity Coefficient (DSC) of 89.60 ± 3.56% and a Hausdorff distance of 1.49 ± 0.65 mm.
2
There was a statistically significant difference in the minimum (Dmin) and maximum (Dmax) doses of radiation received by the CTV when comparing manually planned doses to automatically planned doses.
3
The CNN-based auto-contouring system can be used for segmentation of CTVs to facilitate the brain tumour radiotherapy workflow.

Research Summary

This study evaluated a deep convolutional neural network (CNN) model for the automatic segmentation of the clinical target volume (CTV) in glioblastoma patients, aiming to improve the efficiency and consistency of radiotherapy planning. The modified SegNet model, trained on a large dataset and fine-tuned with independent clinical data, achieved high segmentation accuracy and reduced the need for CT/MRI image registration. The results suggest that the CNN-based auto-contouring system has the potential to facilitate the brain tumour radiotherapy workflow, although manual editing is still necessary for patient safety.

Practical Implications

Improved Radiotherapy Workflow

The CNN-based auto-contouring system can significantly shorten contouring time, leading to a more efficient radiotherapy planning process.

Reduced Inter-Observer Variability

Automated segmentation can mitigate the inconsistencies and biases associated with manual contouring by different radiation oncologists.

Enhanced Treatment Accuracy

By providing a more precise and consistent CTV delineation, the system can potentially lead to more accurate and effective radiation therapy delivery.

Study Limitations

1
The modified SegNet architecture was trained with a small number of subjects.
2
We evaluated deep leaning-based auto-segmented contours on a relatively small number of patients from a single centre.
3
Using a larger cohort from multi-centre datasets may better assess the model’s robustness.

Your Feedback

Was this summary helpful?

Back to Oncology