Deep learning-based 4D-synthetic CTs from sparse-view CBCTs for dose calculations in adaptive proton therapy

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Time-resolved 4D cone beam-computed tomography (4D-CBCT) enables daily anatomical and motion assessment but suffers from artifacts impacting CT number accuracy for proton dose calculations.
• Deep learning offers a solution by correcting CT numbers and generating synthetic CTs (sCTs) to facilitate CBCT-based proton dose calculations.

Purpose of the Study:

• To convert sparse view 4D-CBCTs into 4D-sCTs using a deep convolutional neural network (DCNN).
• To evaluate the image quality and dosimetric accuracy of 4D-sCTs for feasible proton dose calculations in adaptive proton therapy for lung cancer.

Main Methods:

• A DCNN was trained and evaluated on 45 thoracic cancer patients' data, using sparse view 4D-CBCTs reconstructed from 3D acquisition protocols.
• Image quality was assessed using Mean Absolute Error (MAE) and mean error against daily 4D-CTs.
• Dosimetric accuracy was evaluated globally (gamma analysis) and locally for target volumes and organs-at-risk (OARs), including range error simulations and 4D dose reconstruction.

Main Results:

• 4D-sCTs achieved average MAEs of 48.1 ± 6.5 HU (single phase) and 37.7 ± 6.2 HU (average).
• Global dosimetric evaluation showed gamma pass ratios of 92.3% ± 3.2% (single phase) and 94.4% ± 2.1% (average).
• Clinical target volume dose agreement was high (D98 differences < 2.4%), while OAR mean dose differences were up to 8.4%; lung tissues showed higher range errors.

Conclusions:

• Deep learning-based 4D-sCTs demonstrate accuracy for daily dose calculations in adaptive proton therapy.
• Comparable global and local dosimetric accuracy was achieved despite image quality differences between 4D-sCTs and 3D-sCTs.
• Improving CT number accuracy in lung tissues is crucial for further enhancement of 4D lung sCTs.