Masters Degrees (Medical Physics)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/70981

Browse

Browsing Masters Degrees (Medical Physics) by Author "Duprez, Didier Raphael Roger"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    The application of a deep convolution neural network for the automated delineation of the target and organs at risk in High Dose Rate Cervical Brachytherapy
    (Stellenbosch : Stellenbosch University, 2022-12) Duprez, Didier Raphael Roger; Trauernicht, Christoph Jan; Stellenbosch University. Faculty of Medicine and Health Sciences. Dept. of Medical Imaging and Clinical Oncology. Medical Physics.
    ENGLISH SUMMARY: Low/middle income countries suffer from large deficits in experienced Radiation Oncologists, Medical Physicists and Radiation Therapists. Due to these deficits, the bottlenecks experienced in the High-dose rate (HDR) cervical brachytherapy treatment planning workflow are amplified. Image-guided HDR cervical brachytherapy is a complex, labour intensive, manual, time-consuming and expertise driven process. Automation in radiotherapy treatment planning, especially in brachytherapy, has the potential to substantially reduce the overall planning time however most of these algorithms require high level of expertise to develop. The aim of this study is to implement the out of the box self-configuring deep neural network package, known as No New U-Net (nnU-Net), for the task of automatically delineating the organs at risk (OARs) and high-risk clinical target volume (HR CTV) for HDR cervical brachytherapy. The computed tomography (CT) scans of 100 previously treated patients were used to train and test three different nnU-Net configurations (2D, 3DFR and 3DCasc). The performance of the models was evaluated by calculating the Sørensen-Dice similarity coefficient, Hausdorff distance (HD), 95th percentile Hausdorff distance, mean surface distance (MSD) and precision score for 20 test patients. The dosimetric accuracy between the manual and predicted contours was assessed by looking at the various dose volume histogram (DVH) parameters and volume differences. Three different radiation oncologists (ROs) scored the predicted bladder, rectum and HR CTV contours generated by the best performing model. The manual contouring, prediction and editing times were recorded. The mean DSC, HD, HD95, MSD and precision scores for our best performing model (3DFR) were 0.92/7.5 mm/3.0 mm/ 0.8 mm/0.91 for the bladder, 0.84/13.8 mm/5.2 mm/1.4 mm/0.84 for the rectum and 0.81/8.5 mm/6.0 mm/2.2 mm/0.80 for the HR CTV. Mean dose differences (D2cc/90%) and volume differences were 0.08 Gy/1.3 cm3 for the bladder, 0.02 Gy/0.7 cm3 for the rectum and 0.33 Gy/1.5 cm3 for the HR CTV. On average, 65 % of the generated contours were clinically acceptable, 33 % requiring minor edits, 2 % required major edits and no contours were rejected. Average manual contouring time was 14.0 minutes, while the average prediction and editing times were 1.6 and 2.1 minutes respectively. Our best performing model (3DFR) provided fast accurate auto generated OARs and HR CTV contours with a large clinical acceptance rate. Future work should focus on including larger datasets to eliminate inconsistencies, as well as focus on automating the generation of treatment plans.