Search

Purpose: Deep learning‐based networks have become increasingly popular in the field of medical image segmentation. The purpose of this research was to develop and optimize a new architecture for automatic segmentation of the prostate gland and normal organs in the pelvic, thoracic, and upper gastro‐intestinal (GI) regions. Methods: We developed an architecture which combines a shifted‐window (Swin) transformer with a convolutional U‐Net. The network includes a parallel encoder, a cross‐fusion block, and a CNN‐based decoder to extract local and global information and merge related features on the same scale. A skip connection is applied between the cross‐fusion block and decoder to integrate low‐level semantic features. Attention gates (AGs) are integrated within the CNN to suppress features in image background regions. Our network is termed "SwinAttUNet." We optimized the architecture for automatic image segmentation. Training datasets consisted of planning‐CT datasets from 300 prostate cancer patients from an institutional database and 100 CT datasets from a publicly available dataset (CT‐ORG). Images were linearly interpolated and resampled to a spatial resolution of (1.0 × 1.0× 1.5) mm3. A volume patch (192 × 192 × 96) was used for training and inference, and the dataset was split into training (75%), validation (10%), and test (15%) cohorts. Data augmentation transforms were applied consisting of random flip, rotation, and intensity scaling. The loss function comprised Dice and cross‐entropy equally weighted and summed. We evaluated Dice coefficients (DSC), 95th percentile Hausdorff Distances (HD95), and Average Surface Distances (ASD) between results of our network and ground truth data. Results: SwinAttUNet, DSC values were 86.54 ± 1.21, 94.15 ± 1.17, and 87.15 ± 1.68% and HD95 values were 5.06 ± 1.42, 3.16 ± 0.93, and 5.54 ± 1.63 mm for the prostate, bladder, and rectum, respectively. Respective ASD values were 1.45 ± 0.57, 0.82 ± 0.12, and 1.42 ± 0.38 mm. For the lung, liver, kidneys and pelvic bones, respective DSC values were: 97.90 ± 0.80, 96.16 ± 0.76, 93.74 ± 2.25, and 89.31 ± 3.87%. Respective HD95 values were: 5.13 ± 4.11, 2.73 ± 1.19, 2.29 ± 1.47, and 5.31 ± 1.25 mm. Respective ASD values were: 1.88 ± 1.45, 1.78 ± 1.21, 0.71 ± 0.43, and 1.21 ± 1.11 mm. Our network outperformed several existing deep learning approaches using only attention‐based convolutional or Transformer‐based feature strategies, as detailed in the results section. Conclusions: We have demonstrated that our new architecture combining Transformer‐ and convolution‐based features is able to better learn the local and global context for automatic segmentation of multi‐organ, CT‐based anatomy. [ABSTRACT FROM AUTHOR]

Purpose: Task automation is essential for efficient and consistent image segmentation in radiation oncology. We report on a deep learning architecture, comprising a U‐Net and a variational autoencoder (VAE) for automatic contouring of the prostate gland incorporating interobserver variation for radiotherapy treatment planning. The U‐Net/VAE generates an ensemble set of segmentations for each image CT slice. A novel outlier mitigation (OM) technique was implemented to enhance the model segmentation accuracy. Methods: The primary source dataset (source_prim) consisted of 19 200 CT slices (from 300 patient planning CT image datasets) with manually contoured prostate glands. A smaller secondary source dataset (source_sec) comprised 640 CT slices (from 10 patient CT datasets), where prostate glands were segmented by 5 independent physicians on each dataset to account for interobserver variability. Data augmentation via random rotation (<5 degrees), cropping, and horizontal flipping was applied to each dataset to increase sample size by a factor of 100. A probabilistic hierarchical U‐Net with VAE was implemented and pretrained using the augmented source_prim dataset for 30 epochs. Model parameters of the U‐Net/VAE were fine‐tuned using the augmented source_sec dataset for 100 epochs. After the first round of training, outlier contours in the training dataset were automatically detected and replaced by the most accurate contours (based on Dice similarity coefficient, DSC) generated by the model. The U‐Net/OM‐VAE was retrained using the revised training dataset. Metrics for comparison included DSC, Hausdorff distance (HD, mm), normalized cross‐correlation (NCC) coefficient, and center‐of‐mass (COM) distance (mm). Results: Results for U‐Net/OM‐VAE with outliers replaced in the training dataset versus U‐Net/VAE without OM were as follows: DSC = 0.82 ± 0.01 versus 0.80 ± 0.02 (p = 0.019), HD = 9.18 ± 1.22 versus 10.18 ± 1.35 mm (p = 0.043), NCC = 0.59 ± 0.07 versus 0.62 ± 0.06, and COM = 3.36 ± 0.81 versus 4.77 ± 0.96 mm over the average of 15 contours. For the average of 15 highest accuracy contours, values were as follows: DSC = 0.90 ± 0.02 versus 0.85 ± 0.02, HD = 5.47 ± 0.02 versus 7.54 ± 1.36 mm, and COM = 1.03 ± 0.58 versus 1.46 ± 0.68 mm (p < 0.03 for all metrics). Results for the U‐Net/OM‐VAE with outliers removed were as follows: DSC = 0.78 ± 0.01, HD = 10.65 ± 1.95 mm, NCC = 0.46 ± 0.10, COM = 4.17 ± 0.79 mm for the average of 15 contours, and DSC = 0.88 ± 0.02, HD = 7.00 ± 1.17 mm, COM = 1.58 ± 0.63 mm for the average of 15 highest accuracy contours. All metrics for U‐Net/VAE trained on the source_prim and source_sec datasets via pretraining, followed by fine‐tuning, show statistically significant improvement over that trained on the source_sec dataset only. Finally, all metrics for U‐Net/VAE with or without OM showed statistically significant improvement over those for the standard U‐Net. Conclusions: A VAE combined with a hierarchical U‐Net and an OM strategy (U‐Net/OM‐VAE) demonstrates promise toward capturing interobserver variability and produces accurate prostate auto‐contours for radiotherapy planning. The availability of multiple contours for each CT slice enables clinicians to determine trade‐offs in selecting the "best fitting" contour on each CT slice. Mitigation of outlier contours in the training dataset improves prediction accuracy, but one must be wary of reduction in variability in the training dataset. [ABSTRACT FROM AUTHOR]

Purpose: This study introduces an in‐house‐designed software platform (ROdiomiX) for the radiomics analysis of medical images in radiation oncology. ROdiomiX is a MATLAB‐based framework for the computation of radiomic features and feature aggregation techniques in compliance with the Image‐Biomarker‐Standardization‐Initiative (IBSI) guidelines, which includes preprocessing protocols and quantitative benchmark results for analysis of computational phantom images. Methods and Materials: The ROdiomiX software system consists of a series of computation cores implemented on the basis of the guidelines proposed by the IBSI. It is capable of quantitative computation of the following 11 different feature categories: Local‐Intensity, Intensity‐Histogram, Intensity‐Based‐Statistical, Intensity‐Volume‐Histogram, Gray‐Level‐Co‐occurrence, Gray‐Level‐Run‐Length, Gray‐Level‐Size‐Zone, Gray‐Level‐Distance‐Zone, Neighborhood‐Grey‐Tone‐Difference, Neighboring‐Grey‐Level‐Dependence, and Morphological feature. ROdiomiX was validated against benchmark values for the IBSI 3D digital phantom, as well as one designed in‐house (HFH). The intraclass correlation coefficient (ICC) for estimating the degree of absolute agreement between ROdiomiX computation and benchmark values for different features at the 95% confidence level (CL) was used for comparison. Results: Among the 11 feature categories with 149 total features including 10 different feature aggregation methods (following the IBSI guidelines), the percent difference between absolute feature values computed by the ROdiomiX software and benchmark values reported for IBSI and HFH digital phantoms were 0.14% + 0.43% and 0.11% + 0.27%, respectively. The ICC values were >0.997 for all ten feature categories for both the IBSI and HFH digital phantoms. Conclusion: The authors successfully developed a platform for computation of quantitative radiomic features. The image preprocessing and computational software cores were designed following the procedures specified by the IBSI. Benchmarking testing was in excellent agreement against the IBSI‐ and HFH‐designed computational phantoms. [ABSTRACT FROM AUTHOR]