Kuo Men

Automatic segmentation of the clinical target volume and organs at risk in the planning CT for rectal cancer using deep dilated convolutional neural networks

Purpose: Delineation of the clinical target volume (CTV) and organs at risk (OARs) is very important for radiotherapy but is time‐consuming and prone to inter‐observer variation. Here, we proposed a novel deep dilated convolutional neural network (DDCNN)‐based method for fast and consistent auto‐segmentation of these structures. Methods: Our DDCNN method was an end‐to‐end architecture enabling fast training and testing. Specifically, it employed a novel multiple‐scale convolutional architecture to extract multiple‐scale context features in the early layers, which contain the original information on fine texture and boundaries and which are very useful for accurate auto‐segmentation. In addition, it enlarged the receptive fields of dilated convolutions at the end of networks to capture complementary context features. Then, it replaced the fully connected layers with fully convolutional layers to achieve pixel‐wise segmentation. We used data from 278 patients with rectal cancer for evaluation. The CTV and OARs were delineated and validated by senior radiation oncologists in the planning computed tomography (CT) images. A total of 218 patients chosen randomly were used for training, and the remaining 60 for validation. The Dice similarity coefficient (DSC) was used to measure segmentation accuracy. Results: Performance was evaluated on segmentation of the CTV and OARs. In addition, the performance of DDCNN was compared with that of U‐Net. The proposed DDCNN method outperformed the U‐Net for all segmentations, and the average DSC value of DDCNN was 3.8% higher than that of U‐Net. Mean DSC values of DDCNN were 87.7% for the CTV, 93.4% for the bladder, 92.1% for the left femoral head, 92.3% for the right femoral head, 65.3% for the intestine, and 61.8% for the colon. The test time was 45 s per patient for segmentation of all the CTV, bladder, left and right femoral heads, colon, and intestine. We also assessed our approaches and results with those in the literature: our system showed superior performance and faster speed. Conclusions: These data suggest that DDCNN can be used to segment the CTV and OARs accurately and efficiently. It was invariant to the body size, body shape, and age of the patients. DDCNN could improve the consistency of contouring and streamline radiotherapy workflows.

Comparison of setup error using different reference images: a phantom and lung cancer patients study

The purpose of this study was to compare setup errors obtained with kilovoltage cone-beam computed tomography (CBCT) and 2 different kinds of reference images, free-breathing 3D localization CT images (FB-CT) and the average images of 4-D localization CT images (AVG-CT) for phantom and lung cancer patients. This study also explored the correlation between the difference of translational setup errors and the gross tumor volume (GTV) motion. A respiratory phantom and 14 patients were enrolled in this study. For phantom and each patient, 3D helical CT and 4D CT images were acquired, and AVG-CT images were generated from the 4D CT. The setup errors were determined based on the image registration between the CBCT and the 2 different reference images, respectively. The data for both translational and rotational setup errors were analyzed and compared. The GTV centroid movement as well as its correlation with the translational setup error differences was also evaluated. In the phantom study, the AVG-CT method was more accurate than the FB-CT method. For patients, the translational setup errors based on FB-CT were significantly larger than those from AVG-CT in the left-right (LR), superior-inferior (SI), and anterior-posterior (AP) directions (p < 0.05). Translational setup errors differed by >1 mm in 32.6% and >2 mm in 12.9% of CBCT scans. The rotational setup errors from FB-CT were significantly different from those from AVG-CT in the LR and AP directions (p < 0.05). The correlation coefficient of the translational setup error differences and the GTV centroid movement in the LR, SI, and AP directions was 0.515 (p = 0.060), 0.902 (p < 0.001), and 0.510 (p = 0.062), respectively. For lung cancer patients, respiration may affect the on-line target position location. AVG-CT provides different reference information than FB-CT. The difference in SI direction caused by the 2 methods increases with the GTV movement. Therefore, AVG-CT should be the prefered choice of reference images.

Deep Deconvolutional Neural Network for Target Segmentation of Nasopharyngeal Cancer in Planning Computed Tomography Images

Background Radiotherapy is one of the main treatment methods for nasopharyngeal carcinoma (NPC). It requires exact delineation of the nasopharynx gross tumor volume (GTVnx), the metastatic lymph node gross tumor volume (GTVnd), the clinical target volume (CTV), and organs at risk in the planning computed tomography images. However, this task is time-consuming and operator dependent. In the present study, we developed an end-to-end deep deconvolutional neural network (DDNN) for segmentation of these targets. Methods The proposed DDNN is an end-to-end architecture enabling fast training and testing. It consists of two important components: an encoder network and a decoder network. The encoder network was used to extract the visual features of a medical image and the decoder network was used to recover the original resolution by deploying deconvolution. A total of 230 patients diagnosed with NPC stage I or stage II were included in this study. Data from 184 patients were chosen randomly as a training set to adjust the parameters of DDNN, and the remaining 46 patients were the test set to assess the performance of the model. The Dice similarity coefficient (DSC) was used to quantify the segmentation results of the GTVnx, GTVnd, and CTV. In addition, the performance of DDNN was compared with the VGG-16 model. Results The proposed DDNN method outperformed the VGG-16 in all the segmentation. The mean DSC values of DDNN were 80.9% for GTVnx, 62.3% for the GTVnd, and 82.6% for CTV, whereas VGG-16 obtained 72.3, 33.7, and 73.7% for the DSC values, respectively. Conclusion DDNN can be used to segment the GTVnx and CTV accurately. The accuracy for the GTVnd segmentation was relatively low due to the considerable differences in its shape, volume, and location among patients. The accuracy is expected to increase with more training data and combination of MR images. In conclusion, DDNN has the potential to improve the consistency of contouring and streamline radiotherapy workflows, but careful human review and a considerable amount of editing will be required.

Dual-energy imaging method to improve the image quality and the accuracy of dose calculation for cone-beam computed tomography

PURPOSE To improve the image quality and accuracy of dose calculation for cone-beam computed tomography (CT) images through implementation of a dual-energy cone-beam computed tomography method (DE-CBCT), and evaluate the improvement quantitatively. METHODS Two sets of CBCT projections were acquired using the X-ray volumetric imaging (XVI) system on a Synergy (Elekta, Stockholm, Sweden) system with 120kV (high) and 70kV (low) X-rays, respectively. Then, the electron density relative to water (relative electron density (RED)) of each voxel was calculated using a projection-based dual-energy decomposition method. As a comparison, single-energy cone-beam computed tomography (SE-CBCT) was used to calculate RED with the Hounsfield unit-RED calibration curve generated by a CIRS phantom scan with identical imaging parameters. The imaging dose was measured with a dosimetry phantom. The image quality was evaluated quantitatively using a Catphan 503 phantom with the evaluation indices of the reproducibility of the RED values, high-contrast resolution (MTF50%), uniformity, and signal-to-noise ratio (SNR). Dose calculation of two simulated volumetric-modulated arc therapy plans using an Eclipse treatment-planning system (Varian Medical Systems, Palo Alto, CA, USA) was performed on an Alderson Rando Head and Neck (H&N) phantom and a Pelvis phantom. Fan-beam planning CT images for the H&N and Pelvis phantom were set as the reference. A global three-dimensional gamma analysis was used to compare dose distributions with the reference. The average gamma values for targets and OAR were analyzed with paired t-tests between DE-CBCT and SE-CBCT. RESULTS In two scans (H&N scan and body scan), the imaging dose of DE-CBCT increased by 1.0% and decreased by 1.3%. It had a better reproducibility of the RED values (mean bias: 0.03 and 0.07) compared with SE-CBCT (mean bias: 0.13 and 0.16). It also improved the image uniformity (57.5% and 30.1%) and SNR (9.7% and 2.3%), but did not affect the MTF50%. Gamma analyses of the 3D dose distribution with criteria of 1%/1mm showed a pass rate of 99.0-100% and 85.3-97.6% for DE-CBCT and 73.5-99.1% and 80.4-92.7% for SE-CBCT. The average gamma values were reduced significantly by DE-CBCT (p< 0.05). Gamma index maps showed that matching of the dose distribution between CBCT-based and reference was improved by DE-CBCT. CONCLUSIONS DE-CBCT can achieve both better image quality and higher accuracy of dose calculation, and could be applied to adaptive radiotherapy.

Comparison of dosimetric characteristics between stationary and rotational gamma ray stereotactic radiosurgery systems based on Monte Carlo simulation

Stationary and rotational source are two types of source configuration used in commercial Gamma Ray Stereotactic Radiosurgery Systems (GRSRSs). However, it is unclear which one is better in dosimetric performance for clinical use. In this study, precise dose distributions in high resolution of the single source channel of two GRSRSs (the Elekta Leksell Gamma Knife® Model 4C (LGK 4C) for stationary GRSRS and the OUR XGD for rotational GRSRS) were generated by Monte Carlo simulation. Because of the geometrical symmetry, the overall dose distribution is generated by combining the transformed dose distributions of the single source channel. Output factors and penumbra widths were calculated and compared. The differences of output factors between two GRSRSs are minor. The penumbra widths of dose profile from all sources of the LGK 4C are 2.2-1.3-1.4 mm, 3.5-1.6-1.7 mm, 5.8-2.2-2.4 mm, and 7.2-2.6-3.0 mm (in XY-(+Z)-(−Z) form) for the secondary collimator of 4 mm, 8 mm, 14 mm, and 18 mm, respectively and those of the OUR XGD are 3.1-2.3-2.4 mm, 4.0-2.4-2.5 mm, 5.8-2.6-3.0 mm, and 7.2-3.0-3.6 mm for the secondary collimator of 5 mm, 10 mm, 15 mm, and 20 mm, respectively. The dose profiles in Z direction are not symmetrical, because all sources are distributed in +Z direction. Secondly, since the sources are located in higher latitude in the OUR XGD (14°–43°) comparing with that in the LGK 4C (6°–36°), the penumbra widths in Z direction of the OUR XGD are larger than those of the LGK 4C. Thirdly, although the penumbra widths of dose profiles from the single source channel of the LGK 4C are smaller than those of the OUR XGD, with rotational source configuration the OUR XGD has smaller penumbra width of dose profile from all sources in X–Y plane when the secondary collimator is more than 10 mm.

A Method to Improve Electron Density Measurement of Cone-Beam CT Using Dual Energy Technique

Purpose. To develop a dual energy imaging method to improve the accuracy of electron density measurement with a cone-beam CT (CBCT) device. Materials and Methods. The imaging system is the XVI CBCT system on Elekta Synergy linac. Projection data were acquired with the high and low energy X-ray, respectively, to set up a basis material decomposition model. Virtual phantom simulation and phantoms experiments were carried out for quantitative evaluation of the method. Phantoms were also scanned twice with the high and low energy X-ray, respectively. The data were decomposed into projections of the two basis material coefficients according to the model set up earlier. The two sets of decomposed projections were used to reconstruct CBCT images of the basis material coefficients. Then, the images of electron densities were calculated with these CBCT images. Results. The difference between the calculated and theoretical values was within 2% and the correlation coefficient of them was about 1.0. The dual energy imaging method obtained more accurate electron density values and reduced the beam hardening artifacts obviously. Conclusion. A novel dual energy CBCT imaging method to calculate the electron densities was developed. It can acquire more accurate values and provide a platform potentially for dose calculation.

Fully automatic and robust segmentation of the clinical target volume for radiotherapy of breast cancer using big data and deep learning

PURPOSE To train and evaluate a very deep dilated residual network (DD-ResNet) for fast and consistent auto-segmentation of the clinical target volume (CTV) for breast cancer (BC) radiotherapy with big data. METHODS DD-ResNet was an end-to-end model enabling fast training and testing. We used big data comprising 800 patients who underwent breast-conserving therapy for evaluation. The CTV were validated by experienced radiation oncologists. We performed a fivefold cross-validation to test the performance of the model. The segmentation accuracy was quantified by the Dice similarity coefficient (DSC) and the Hausdorff distance (HD). The performance of the proposed model was evaluated against two different deep learning models: deep dilated convolutional neural network (DDCNN) and deep deconvolutional neural network (DDNN). RESULTS Mean DSC values of DD-ResNet (0.91 and 0.91) were higher than the other two networks (DDCNN: 0.85 and 0.85; DDNN: 0.88 and 0.87) for both right-sided and left-sided BC. It also has smaller mean HD values of 10.5 mm and 10.7 mm compared with DDCNN (15.1 mm and 15.6 mm) and DDNN (13.5 mm and 14.1 mm). Mean segmentation time was 4 s, 21 s and 15 s per patient with DDCNN, DDNN and DD-ResNet, respectively. The DD-ResNet was also superior with regard to results in the literature. CONCLUSIONS The proposed method could segment the CTV accurately with acceptable time consumption. It was invariant to the body size and shape of patients and could improve the consistency of target delineation and streamline radiotherapy workflows.

A feasibility study on an automated method to generate patient-specific dose distributions for radiotherapy using deep learning

Purpose To develop a method for predicting optimal dose distributions, given the planning image and segmented anatomy, by applying deep learning techniques to a database of previously optimized and approved Intensity‐modulated radiation therapy treatment plans. Methods Eighty cases of early‐stage nasopharyngeal cancer (NPC) were included in the study. Seventy cases were chosen randomly as the training set and the remaining as the test set. The inputs were the images with structures, with each target and organs at risk (OARs) assigned a unique label. The outputs were dose maps, including coarse dose maps and converted fine dose maps (FDM) from convolution. Two types of input images with structures were used in the model building. One type of input included the images (with associated structures) without manipulation. The second type of input involved modifying the image gray label with information from radiation beam geometry. ResNet101 was chosen as the deep learning network for both. The accuracy of predicted dose distributions was evaluated against the corresponding dose as used in the clinic. A global three‐dimensional gamma analysis was calculated for the evaluation. Results The proposed model trained with the two different sets of input images and structures could both predict patient‐specific dose distributions accurately. For the out‐of‐field dose distributions, the model obtained from the input with radiation geometry performed better (dose difference in %, 4.7 ± 6.1% vs 5.5 ± 7.9%, P < 0.05). The mean Gamma pass rates of dose distributions predicted with both types of input were comparable for most OARs (P > 0.05), except for the bilateral optic nerves and the optic chiasm. Conclusions The proposed system with radiation geometry added to the input is a promising method to generate patient‐specific dose distributions for radiotherapy. It can be applied to obtain the dose distributions slice‐by‐slice for planning quality assurance and for guiding automated planning.

Influence of tumor location on the intensity-modulated radiation therapy plan of helical tomotherapy

Given the design of the Helical TomoTherapy device, the patients central axis is routinely aligned with the machines rotational axis to prevent the patients body from colliding with the machine walls. However, for treatment of tumors located away from the patients central axis, this position may not be optimal as the adequate radiation dose may not reach the affected site. Our study aimed to investigate the influence of tumor location on dose quality and delivery efficiency of tomotherapy plans. A phantom and 15 patients were selected for this study. Two plans, A and B, were implemented for each case. In plan A, the patients central axis was aligned with the machines rotational axis, whereas in plan B, the center of the planning target volume (PTV) was aligned with the machines rotational axis. Both plans were optimized with the same planning parameters, and the dose quality of the plans was evaluated using dosimetrics. The delivery efficiency was determined from delivery time and monitor units (MUs). A paired t-test or nonparametric Wilcoxon signed-rank test was performed for statistical comparison. In the phantom study, the median delivery times were 358 and 336 seconds for plans A and B, respectively, and this difference was significant (p = 0.005). In the patient study, the median delivery times were 348 and 317 seconds for plans A and B, respectively, and this difference was also significant (p = 0.001). The dose qualities of both plans for each patient were nearly identical. No significant differences were found in the conformal index, heterogeneity index, and mean dose delivered to normal tissue between the plans. Both phantom and patient studies showed that for normal-sized patients, the delivery time reduced as the distance between the PTV and the patients central axis increased when the PTV center was aligned with the machine axis. In conclusion, aligning the PTV center with the machines rotational axis by shifting the patient during tomotherapy reduces the delivery time without compromising the dose quality of intensity-modulated radiation therapy.