Deep learning based automatic internal gross target volume delineation from 4D-CT of hepatocellular carcinoma patients.

BACKGROUND: The location and morphology of the liver are significantly affected by respiratory motion. Therefore, delineating the gross target volume (GTV) based on 4D medical images is more accurate than regular 3D-CT with contrast. However, the 4D method is also more time-consuming and laborious. This study proposes a deep learning (DL) framework based on 4D-CT that can achieve automatic delineation of internal GTV. METHODS: The proposed network consists of two encoding paths, one for feature extraction of adjacent slices (spatial slices) in a specific 3D-CT sequence, and one for feature extraction of slices at the same location in three adjacent phase 3D-CT sequences (temporal slices), a feature fusion module based on an attention mechanism was proposed for fusing the temporal and spatial features. Twenty-six patients' 4D-CT, each consisting of 10 respiratory phases, were used as the dataset. The Hausdorff distance (HD95), Dice similarity coefficient (DSC), and volume difference (VD) between the manual and predicted tumor contour were computed to evaluate the model's segmentation accuracy. RESULTS: The predicted GTVs and IGTVs were compared quantitatively and visually with the ground truth. For the test dataset, the proposed method achieved a mean DSC of 0.869 ± 0.089 and an HD95 of 5.14 ± 3.34 mm for all GTVs, with under-segmented GTVs on some CT slices being compensated by GTVs on other slices, resulting in better agreement between the predicted IGTVs and the ground truth, with a mean DSC of 0.882 ± 0.085 and an HD95 of 4.88 ± 2.84 mm. The best GTV results were generally observed at the end-inspiration stage. CONCLUSIONS: Our proposed DL framework for tumor segmentation on 4D-CT datasets shows promise for fully automated delineation in the future. The promising results of this work provide impetus for its integration into the 4DCT treatment planning workflow to improve hepatocellular carcinoma radiotherapy.

Deep learning-based internal gross target volume definition in 4D CT images of lung cancer patients.

BACKGROUND: Contouring of internal gross target volume (iGTV) is an essential part of treatment planning in radiotherapy to mitigate the impact of intra-fractional target motion. However, it is usually time-consuming and easily subjected to intra-observer and inter-observer variability. So far, few studies have been explored to directly predict iGTV by deep learning technique, because the iGTV contains not only the gross target volume (GTV) but also the motion information of the GTV. PURPOSE: This work was an exploratory study to present a deep learning-based framework to segment iGTV rapidly and accurately in 4D CT images for lung cancers. METHODS: Five models, including 3D UNet, mmUNet with point-wise add merging approach (mmUNet-add), mmUNet with concatenate fusion strategy (mmUNet-cat), gruUNet with point-wise add fusion approach (gruUNet-add), and gruUNet with concatenate method (gruUNet-cat), were adopted for iGTV segmentation. All the models originated from the 3D UNet network, with multi-channel multi-path and convolutional gated recurrent unit (GRU) added in the mmUNet and gruUNet networks, respectively. Seventy patients with lung cancers were collected and 55 cases were randomly selected as the training set, and 15 cases as the testing set. In addition, the segmentation results of the five models were compared with the ground truths qualitatively and quantitatively. RESULTS: In terms of Dice Similarity Coefficient (DSC), the proposed four networks (mmUNet-add, mmUNet-cat, gruUNet-add, and gruUNet-cat) increased the DSC score of 3D UNet from 0.6945 to 0.7342, 0.7253, 0.7405, and 0.7365, respectively. However, the differences were not statistically significant (p > 0.05). After a simple post-processing to remove the small isolated connected regions, the mean 95th percentile Hausdorff distances (HD_95s) of the 3D UNet, mmUNet-add, mmUNet-cat, gruUNet-add, and gruUNet-cat networks were 19.70, 15.75, 15.84, 15.61, and 15.83 mm, respectively, corresponding to 25.35, 25.96, 25.11, 28.23, and 24.47 mm before the post-processing. With regard to runtime, significant elapsed time growths (about 70s and 230s) were observed both in the mmUNet and gruUNet architectures due to the increasing parameters. But the mmUNet structure showed less growth. CONCLUSION: Our study demonstrated the ability of the deep learning technique to predict iGTVs directly. With the introduction of multi-channel multi-path and convolutional GRU, the segmentation accuracy was improved under certain conditions with a reduced segmentation efficiency and a further research topic when the 3D UNet network would lead to poor performance is elicited. Less efficiency degradation was observed in the mmUNet structure. Besides, the element-wise add fusing strategy was favorable to increase DSC, whereas HD_95 benefited from the concentrate merging approach. Nevertheless, the segmentation accuracy by deep learning still remains to be improved.

Comparison of primary tumour volumes delineated on four-dimensional computed tomography maximum intensity projection and (18) F-fluorodeoxyglucose positron emission tomography computed tomography images of non-small cell lung cancer.

INTRODUCTION: The study aims to compare the positional and volumetric differences of tumour volumes based on the maximum intensity projection (MIP) of four-dimensional CT (4DCT) and (18) F-fluorodexyglucose ((18) F-FDG) positron emission tomography CT (PET/CT) images for the primary tumour of non-small cell lung cancer (NSCLC). METHODS: Ten patients with NSCLC underwent 4DCT and (18) F-FDG PET/CT scans of the thorax on the same day. Internal gross target volumes (IGTVs) of the primary tumours were contoured on the MIP images of 4DCT to generate IGTVMIP . Gross target volumes (GTVs) based on PET (GTVPET ) were determined with nine different threshold methods using the auto-contouring function. The differences in the volume, position, matching index (MI) and degree of inclusion (DI) of the GTVPET and IGTVMIP were investigated. RESULTS: In volume terms, GTVPET 2.0 and GTVPET 20% approximated closely to IGTVMIP with mean volume ratio of 0.93 ± 0.45 and 1.06 ± 0.43, respectively. The best MI was between IGTVMIP and GTVPET 20% (0.45 ± 0.23). The best DI of IGTVMIP in GTVPET was IGTVMIP in GTVPET 20% (0.61 ± 0.26). CONCLUSIONS: In 3D PET images, the GTVPET contoured by standardised uptake value (SUV) 2.0 or 20% of maximal SUV (SUVmax ) approximate closely to the IGTVMIP in target size, while the spatial mismatch is apparent between them. Therefore, neither of them could replace IGTVMIP in spatial position and form. The advent of 4D PET/CT may improve the accuracy of contouring the perimeter for moving targets.

Detection of the correlations in target volumes for primary tumor of non-small cell lung cancer based on FDG PET/CT and the end-expiration phase of 4D-CT / 中华放射医学与防护杂志

Objective To investigate the correlations in target volumes based on positron emission tomography CT (PET/CT) and the end-expiration phase of four-dimensional CT (4D-CT) images for non-small cell lung cancer (NSCLC).Methods Seventeen patients with NSCLC sequentially underwent three-dimensional CT (3DCT),4D-CT and 18F-FDG PET/CT thoracic simulation scans.The gross target volume (GTV) was contoured on the end-expiration phase (50％) of 4D-CT and defined as GTV50％.The internal gross target volumes (IGTV) based on PET/CT images (IGTVPET) were determined by the standardized uptake value (SUV) 2.0 (IGTVPET2.0) and 20％ percentage of the maximal standardized uptake value (SUVmax) (IGTVPET20％).The following parameters were calculated to analyze the correlation between IGTVPET and GTV50％ in volume ratio (VR) and conformity index (CI):maximum transverse diameter of GTV50％,volume of GTV50％,the displacement of GTV in the cranial-caudal direction and 3D Vector calculated from 4D-CT dataset as well as the SUVmax.Results There was no significant correlation between the VR of IGTVPET2.0 to GTV50％ and the maximum transverse diameter of GTV50％,volume of GTV50％,the displacement of GTV in the cranial-caudal direction,3D Vector and the SUVmax (P ＞ 0.05).The VR between IGTVPET20％ and GTV50％ inversely related to maximum transverse diameter of GTV50％,volume of GTV50％ and SUVmax (r =-0.663,-0.669,-0.752,P ＜0.05).The CI between IGTVPET2.0 and GTV50％ positively related to volume of GTV50％ and maximum transverse diameter of GTV50％ (r =0.613,0.483,P ＜ 0.05).Conclusions 3D PET images provide a time-averaged image of the tumor during the numerous breathing cycle.They fail to include the full information of moving tumor.The target volumes based on 3D PET might not reflect the real IGTV of NSCLC.

Geometrical differences in target volumes based on 18F-fluorodeoxyglucose positron emission tomography/computed tomography and four-dimensional computed tomography maximum intensity projection images of primary thoracic esophageal cancer.

The objective of the study was to compare geometrical differences of target volumes based on four-dimensional computed tomography (4DCT) maximum intensity projection (MIP) and 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT) images of primary thoracic esophageal cancer for radiation treatment. Twenty-one patients with thoracic esophageal cancer sequentially underwent contrast-enhanced three-dimensional computed tomography (3DCT), 4DCT, and 18F-FDG PET/CT thoracic simulation scans during normal free breathing. The internal gross target volume defined as IGTVMIP was obtained by contouring on MIP images. The gross target volumes based on PET/CT images (GTVPET ) were determined with nine different standardized uptake value (SUV) thresholds and manual contouring: SUV≥2.0, 2.5, 3.0, 3.5 (SUVn); ≥20%, 25%, 30%, 35%, 40% of the maximum (percentages of SUVmax, SUVn%). The differences in volume ratio (VR), conformity index (CI), and degree of inclusion (DI) between IGTVMIP and GTVPET were investigated. The mean centroid distance between GTVPET and IGTVMIP ranged from 4.98 mm to 6.53 mm. The VR ranged from 0.37 to 1.34, being significantly (P<0.05) closest to 1 at SUV2.5 (0.94), SUV20% (1.07), or manual contouring (1.10). The mean CI ranged from 0.34 to 0.58, being significantly closest to 1 (P<0.05) at SUV2.0 (0.55), SUV2.5 (0.56), SUV20% (0.56), SUV25% (0.53), or manual contouring (0.58). The mean DI of GTVPET in IGTVMIP ranged from 0.61 to 0.91, and the mean DI of IGTVMIP in GTVPET ranged from 0.34 to 0.86. The SUV threshold setting of SUV2.5, SUV20% or manual contouring yields the best tumor VR and CI with internal-gross target volume contoured on MIP of 4DCT dataset, but 3DPET/CT and 4DCT MIP could not replace each other for motion encompassing target volume delineation for radiation treatment.

Comparison of internal gross target volumes delineated on the maximum intensity projection of four-dimensional CT images and positron emission tomography-CT for primary thoracic esophageal cancer / 中华放射医学与防护杂志

Objective To compare volumetric size, conformity index (CI), degree of inclusion (DI) of internal gross target volumes (IGTV) delineated on 4D-CT-MIP and PET-CT images for primary thoracic esophageal cancer. Methods Fifteen patients with thoracic esophageal cancer sequentially underwent enhanced 3D-CT, 4D-CT and PET-CT simulation scans. IGTVMIP was obtained by contouring on 4D-CT maximum intensity projection ( MIP). The PET contours were determined with nine different threshold methods (SUV≥2?0, 2?5, 3?0, 3?5), the percentages of the SUVmax(≥20%, 25%, 30%, 35%, 40%) and manual contours. The differences in size, conformity index (CI), degree of inclusion ( DI) of different volumes were compared. Results The volume ratios ( VRs) of IGTVPET2. 5 to IGTVMIP , IGTVPET20% to IGTVMIP, IGTVPETMAN to IGTVMIP were 0?86, 0?88, 1?06, respectively, which approached closest to 1. The CIs of IGTVPET2?0,IGTVPET2.5,IGTVPET20%,IGTVPETMAN and IGTVMIP which were 0?55, 0?56, 0?56, 0?54,0?55, respectively, were significantly larger than other CIs of IGTVPET and IGTVMIP (Z= -3?408-2?215,P 0?05). Conclusions The targets delineated based on SUV threshold setting of≥2?5, 20% of the SUVmax and manual contours on PET images correspond better with the target delineated on maximum intensity projection of 4D-CT images than other SUV thresholding methods.