Abstract:Quantitative Cone Beam CT (CBCT) is the foundation for image-guided radiation therapy (RT) techniques. In addition to the current application of patient setup during pancreatic RT, CBCT has the potential to be further employed to assess inter-fraction variability in organs-at-risk (OARs) caused by patient motion, weight loss, and other factors. Accurate segmentations of OARs with on-board CBCT could help online dose evaluation for avoiding severe gastrointestinal toxicity given varying proximity of sensitive OARs to high dose regions; however manual segmentation on daily CBCTs is labor-intensive. Image quality of CBCT is usually far inferior to simulation CT scans, making automated segmentation extremely difficult. This study aims to develop deep-learning-based CBCT segmentation aided by synthetic CT (sCT) to facilitate accurate, automated OAR delineation during CBCT-guided adaptive RT. Forty patients who received RT for pancreatic carcinoma had their OARs contoured on a CBCT scan from their RT. A multi-task deep learning-based method was used to first enhance the CBCT image quality and then simultaneously segment multiple organs. To enhance the CBCT image quality, a cycle-consistent generative adversarial network (CycleGAN) was used to generate high-quality sCT from CBCT. The aligned CT image was used as learning target of CBCT image. To learn the specific difference between CBCT and CT and thus reduce the scatter artifact, nine residual blocks were integrated into the CycleGAN. A deep attention fully convolution network took sCT as input, and output the multi-channel binary mask of OARs, each channel denoted one OAR. Deep attention strategy was used via involved attention gates to automatically highlight the significant features that accurately represent the different organs. We performed five-fold cross-validation experiments to evaluate our method. Dice similarity coefficient (DSC), mean surface distance (MSD) and percentage volume difference (PVD) were used to quantify the differences between our segmentation and manual contour (ground truth). The averaged DSC, MSD and PVD are 0.80±0.20, 1.95±2.37mm and 10.81±9.28% for duodenum, 0.87±0.05, 1.33±0.58mm and 4.06±4.43% for small bowel, 0.88±0.07, 2.10±2.21mm and 4.68±6.16% for large bowel, 0.91±0.08, 1.59±1.92 mm and 6.09±7.37% for stomach, 0.93±0.05, 2.90±4.77 mm and 4.31±5.68% for liver, 0.92±0.05, 2.05±4.71 mm and 5.20±5.35% for left kidney, 0.93±0.04, 2.09±3.40mm and 2.58±5.06% for right kidney and 0.88±0.07, 0.75±0.50 mm and 7.20±9.58% for spinal cord. All OARs were segmented within several seconds. This study demonstrated that a novel sCT-aided deep learning method for multi-organ segmentation is capable of reliably and rapidly segmenting multiple OARs for adaptive RT of pancreatic cancer. The accurate sCT-based pancreatic OAR delineation warrants further development of a CBCT-guided adaptive RT workflow for pancreatic cancer.

Automatic Segmentation of Organs-At-Risk and Clinical Target Volume for Cervical Cancer Using Manifold Learning

Combinatorial CNN-Transformer Learning with Manifold Constraints for Semi-supervised Medical Image Segmentation

Automated Clinical Target Volume Delineation Using Deep 3D Neural Networks in Radiation Therapy of Non-small Cell Lung Cancer

A manifold learning regularization approach to enhance 3D CT image-based lung nodule classification

Simultaneous reconstruction and segmentation of MRI image by manifold learning

Organ at Risk Segmentation in Head and Neck CT Images Using a Two-Stage Segmentation Framework Based on 3D U-Net

Dual convolution-transformer UNet (DCT-UNet) for organs at risk and clinical target volume segmentation in MRI for cervical cancer brachytherapy

Manifold-Aware Local Feature Modeling for Semi-Supervised Medical Image Segmentation

An Encoder-Decoder Network for Automatic Clinical Target Volume Target Segmentation of Cervical Cancer in CT Images

Organ-at-Risk (OAR) Segmentation in Head and Neck CT Using U-RCNN

A deep learning-based framework for segmenting invisible clinical target volumes with estimated uncertainties for post-operative prostate cancer radiotherapy

Segmentation of the prostate and organs at risk in male pelvic CT images using deep learning

Multi-organ segmentation of organ-at-risk (OAR's) of head and neck site using ensemble learning technique

Automatic Segmentation of Head-Neck Organs by Multi-mode CNNs for Radiation Therapy

3D Lightweight Network for Simultaneous Registration and Segmentation of Organs-at-Risk in CT Images of Head and Neck Cancer

Segmentation of Clinical Target Volume From CT Images for Cervical Cancer Using Deep Learning

Synthetic CT-aided Online CBCT Multi-Organ Segmentation for CBCT-guided Adaptive Radiotherapy of Pancreatic Cancer

Multi-organ Segmentation in Head and Neck MRI Using U-Faster-RCNN.

Weaving attention U‐net: A novel hybrid CNN and attention‐based method for organs‐at‐risk segmentation in head and neck CT images

Progressive auto-segmentation for cone-beam computed tomography-based online adaptive radiotherapy

Automatic segmentation of the clinical target volume and organs at risk for rectal cancer radiotherapy using structure-contextual representations based on 3D high-resolution network