Browsing by Author "Chang, Yushi"
- Results Per Page
- Sort Options
Item Open Access An Investigation of Machine Learning Methods for Delta-radiomic Feature Analysis(2018) Chang, YushiBackground: Radiomics is a process of converting medical images into high-dimensional quantitative features and the subsequent mining these features for providing decision support. It is conducted as a potential noninvasive, low-cost, and patient-specific routine clinical tool. Building a predictive model which is reliable, efficient, and accurate is a vital part for the success of radiomics. Machine learning method is a powerful tool to achieve this. Feature extraction strongly affects the performance. Delta-feature is one way of feature extraction methods to reflect the spatial variation in tumor phenotype, hence it could provide better treatment-specific assessment.
Purpose: To compare the performance of using pre-treatment features and delta-features for assessing the brain radiosurgery treatment response, and to investigate the performance of different combinations of machine learning methods for feature selection and for feature classification.
Materials and Methods: A cohort of 12 patients with brain treated by radiosurgery was included in this research. The pre-treatment, one-week post-treatment, and two-month post-treatment T1 and T2 FLAIR MR images were acquired. 61 radiomic features were extracted from the gross tumor volume (GTV) of each image. The delta-features from pre-treatment to two post-treatment time points were calculated. With leave-one-out sampling, pre-treatment features and the two sets of delta-features were separately input into a univariate Cox regression model and a machine learning model (L1-regularized logistic regression [L1-LR], random forest [RF] or neural network [NN]) for feature selection. Then a machine learning method (L1-LR, L2-regularized logistic regression [L2-LR], RF, NN, kernel support vector machine [Kernel-SVM], linear-SVM, or naïve bayes [NB]) was used to build a classification model to predict overall survival. The performance of each model combination and feature type was estimated by the area under receiver operating characteristic (ROC) curve (AUC).
Results: The AUC of one-week delta-features was significantly higher than that of pre-treatment features (p-values < 0.0001) and two-month delta-features (p-value= 0.000). The model combinations of L1-LR for feature selection and RF for classification as well as RF for feature selection and NB for classification based on one-week delta-features presented the highest AUC values (both AUC=0.944).
Conclusions: This work potentially implied that the delta-features could be better in predicting treatment response than pre-treatment features, and the time point of computing the delta-features was a vital factor in assessment performance. Analyzing delta-features using a suitable machine learning approach is potentially a powerful tool for assessing treatment response.
Item Open Access An investigation of machine learning methods in delta-radiomics feature analysis.(PloS one, 2019-01) Chang, Yushi; Lafata, Kyle; Sun, Wenzheng; Wang, Chunhao; Chang, Zheng; Kirkpatrick, John P; Yin, Fang-FangPURPOSE:This study aimed to investigate the effectiveness of using delta-radiomics to predict overall survival (OS) for patients with recurrent malignant gliomas treated by concurrent stereotactic radiosurgery and bevacizumab, and to investigate the effectiveness of machine learning methods for delta-radiomics feature selection and building classification models. METHODS:The pre-treatment, one-week post-treatment, and two-month post-treatment T1 and T2 fluid-attenuated inversion recovery (FLAIR) MRI were acquired. 61 radiomic features (intensity histogram-based, morphological, and texture features) were extracted from the gross tumor volume in each image. Delta-radiomics were calculated between the pre-treatment and post-treatment features. Univariate Cox regression and 3 multivariate machine learning methods (L1-regularized logistic regression [L1-LR], random forest [RF] or neural networks [NN]) were used to select a reduced number of features, and 7 machine learning methods (L1-LR, L2-LR, RF, NN, kernel support vector machine [KSVM], linear support vector machine [LSVM], or naïve bayes [NB]) was used to build classification models for predicting OS. The performances of the total 21 model combinations built based on single-time-point radiomics (pre-treatment, one-week post-treatment, and two-month post-treatment) and delta-radiomics were evaluated by the area under the receiver operating characteristic curve (AUC). RESULTS:For a small cohort of 12 patients, delta-radiomics resulted in significantly higher AUC than pre-treatment radiomics (p-value<0.01). One-week/two-month delta-features resulted in significantly higher AUC (p-value<0.01) than the one-week/two-month post-treatment features, respectively. 18/21 model combinations were with higher AUC from one-week delta-features than two-month delta-features. With one-week delta-features, RF feature selector + KSVM classifier and RF feature selector + NN classifier showed the highest AUC of 0.889. CONCLUSIONS:The results indicated that delta-features could potentially provide better treatment assessment than single-time-point features. The treatment assessment is substantially affected by the time point for computing the delta-features and the combination of machine learning methods for feature selection and classification.Item Open Access Development of an eXtended Modular ANthropomorphic (XMAN) phantom for Imaging and Treatment Optimization in Radiotherapy(2021) Chang, YushiDeveloping new technologies for clinical usage requires systematic and rigorous validation and optimization for various patient scenarios to ensure robustness and accuracy. Clinical trials on real patients are always considered the gold standard for evaluating newly-developed technologies. However, it is often challenging to achieve due to the high cost, difficulties in patient recruitment, ethical issues especially considering the extra imaging dose, lack of ground-truth patient data, insufficient number of patients for data-demanding studies, etc. As a result, anthropomorphic phantoms are utilized to substitute patients for the testing, evaluation, and optimization of various novel radiotherapy and medical imaging techniques before the patient studies. Physical phantoms can be used as a preliminary tool for evaluating some techniques, but they tend to be expensive, over-simplified, and unflexible to simulate various patient scenarios. By contrast, digital phantoms are cost-effective, providing a ground truth with known anatomy and physiological functions, being flexible in simulating different patient scenarios, and efficiently generating a large amount of data. Thus, digital phantoms are more likely to be a powerful tool for diverse technique assessments and substitute for patients in virtual clinical trials.The 4D-extended cardiac-torso (XCAT) phantom is a digital phantom that simulates diverse human anatomies (e.g., genders, heights, tumor sizes etc.) with cardiac and respiratory motions, and it has been widely used in medical imaging and radiotherapy research. Although being widespread, the 4D-XCAT phantom has several limitations. First, it lacks intra-organ anatomical details and textures. The absence of the heterogeneous anatomical textures would degrade its translatability from phantom to patient studies for various research in radiation therapy, such as imaging reconstruction, image registration, segmentation, treatment planning, etc. Second, the breathing motion pattern in the 4D-XCAT phantom is simplified to an approximate linear model. Therefore, the current phantom is not adequate for research that involves lung function, motion management, 4D imaging or treatment optimization, etc. Third, the XCAT phantom has not included onboard imaging artifacts. As a result, the current phantom is not adequate for evaluating treatment delivery techniques related to motion management, onboard localization, etc. This dissertation aims to address these limitations by developing an eXtended Modular ANthropomorphic (XMAN) phantom based on 4D-XCAT using generative adversarial network (GAN)-based deep learning techniques. Structure-wise, this dissertation includes the first chapter for introduction, the second chapter for specific aims addressing the three limitations of the 4D-XCAT phantom, and three chapters (chapters 3, 4, and 5) with one limitation discussed at each one of them. Finally, the dissertation work is summarized in chapter 6. In chapter 3, we proposed to synthesize anthropomorphic anatomical textures in the chest region of the XCAT phantom. A dual-discriminator conditional-GAN (D-CGAN) model was developed. For training purposes, we generated organ maps to mimic XCAT images and train the model to synthesize realistic anatomical textures in the organ maps to simulate real CT or CBCT images. The organ maps were generated by segmenting the organs and gross tumor volumes from the actual patient images and assigning unique HU values to each structure. The D-CGAN was uniquely designed with two discriminators, one trained for the body and the other for the tumor. The D-CGAN model was trained separately using 62 CT and 63 CBCT images from lung SBRT patients. Then, various XCAT phantoms were input to the D-CGAN model to generate multi-contrast textured XCAT (MT-XCAT), including CT and CBCT contrast. The MT-XCAT phantoms were evaluated by comparing the intensity histogram features. The visual examination demonstrated that the MT-XCAT phantoms presented similar general contrast and anatomical textures as CT and CBCT images. The mean HU of the MT-XCAT-CT and MT-XCAT-CBCT were -140.35 ± 336.48 and -185.72 ± 350.32, compared with that of real CT (-149.79 ± 346.37) and CBCT ((-245.41 ± 371.66). The study demonstrated that realistic MT-XCAT phantoms were generated using the proposed method. In chapter 4, we proposed to synthesize realistic and controllable respiratory motions in the XCAT phantoms by developing a GAN-based deep learning technique. A motion generation model was developed based on BicycleGAN with a novel 4D generator. Input with the end-of-inhale (EOI) phase images and a Gaussian perturbation, the model generates inter-phase deformable-vector-fields (DVFs), which are composed and applied to the input to generate 4D images. The model was trained and validated using 71 4D-CT images from lung cancer patients. During testing, the EOI phase images of the 4D-XCAT phantom are input to the well-trained motion generation model to generate 4D-XCAT with realistic respiratory motions. By tuning the input Gaussian perturbation, 4D-XCAT phantoms with different breathing amplitudes are generated. A separate respiratory motion amplitude control model was built using decision tree regression to predict the input perturbation needed for a specific motion amplitude. This model was developed using 300 4D-XCAT phantoms generated from 6 original 4D-XCAT phantoms of different sizes with 50 different perturbations for each size. For evaluating the generated 4D images, Dice coefficients for lungs and lung volume variation during respiration were compared between the simulated images and reference images. Deformation energy was calculated to evaluate the generated DVF. DVFs and ventilation maps of the simulated 4D-CT were compared with the reference 4D-CTs using cross-correlation and Spearman’s correlation. Additionally, a comparison of DVFs and ventilation maps among the original 4D-XCAT, the generated 4D-XCAT, and reference patient 4D-CTs were made to show the improvement of motion realism by the model. The amplitude control error was calculated as the absolute error between the generated and desired breathing amplitude. Results show that the maximum deviation of lung volume during respiration was 5.8% and the Dice coefficient for lungs reached at least 0.95 by comparing the simulated and reference 4D-CTs. The generated DVFs comparable deformation energy levels. The cross-correlation of DVFs achieved 0.89 ± 0.10/ 0.86 ± 0.12/ 0.95 ± 0.04 along the x/ y/ z directions in the testing patient group. The cross-correlation of ventilation maps derived achieved 0.80 ± 0.05/ 0.67 ± 0.09/ 0.68 ± 0.13, and the Spearman's correlation achieved 0.70 ± 0.05/ 0.60 ± 0.09/ 0.53 ± 0.01, respectively, in the training/validation/testing patient groups. The generated 4D-XCAT phantoms presented similar deformation energy as patient data while maintaining the original XCAT phantom (Dice=0.95, maximum lung volume variation=4%). The respiratory motion amplitude error was controlled to be less than 0.5 mm. The results demonstrated that realistic and controllable respiratory motion in the 4D-XCAT phantom was synthesized using the proposed method. In chapter 5, we proposed to synthesize realistic CBCT imaging artifacts in the XCAT phantoms. An artifact simulation model is developed using CT images and image acquisition conditions as inputs and is trained to generate CBCT images with the conditions-specific imaging artifacts. Then, XCAT phantoms are input to the model with user-desired image acquisition conditions to simulate CBCT artifacts in the XCAT phantoms. The model was trained in a two-step scheme: (1) simulate cone and under-sampling artifacts for different projection numbers (450, 150, or 100); (2) simulate scatter and beam-hardening artifacts. Cone-beam projections of the CT volumes were generated by ray-tracing and used to reconstruct CBCTs using FDK-backprojection. These CBCT volumes were used as the output for the first-step model and the input for second-step model. The second-step model outputs corresponding CBCTs reconstructed using Monte-Carlo (MC) projections. The model was trained and validated with 14 patients with 10-fold cross-validation and tested by one independent patient and 5 XCAT phantoms. In addition to qualitative evaluation, we quantitatively evaluated the model performance by peak-signal-to-noise-ratio (PSNR), structural similarity index (SSIM), cross-correlation, and NPS (noise power spectrum) correlation coefficient (NCC). For the under-sampling and cone artifacts simulation, the PSNR reached 38.58/37.24/36.92, and the SSIM reached 0.987/0.978/0.966 for the testing patient with 450/225/100 projections, respectively. The cross-correlation and NCC achieved at least 0.99 for all projection numbers. Qualitatively, the model successfully simulated the streak artifact due to under-sampling and cone artifact in the XCAT phantom. For scatter simulation, the PSNR reached 28.05, and SSIM reached 0.904 for the testing patient. The cross-correlation was 0.927, and NCC was 0.984. The results demonstrated the feasibility of synthesizing CBCT artifacts in the 4D-XCAT phantom using the proposed method. In summary, the results demonstrate that an eXtended Modular ANthropomorphic (XMAN) phantom has been built based on the 4D-XCAT phantom with the following important featurs: (1) Realistic anthropomorphic anatomical textures for CT and CBCT in the chest region. This feature can significantly enhance the translatability from phantom to patient studies, better preparing the phantoms for a wide variety of virtual clinical trials in radiation therapy, such as imaging reconstruction, image registration, segmentation, treatment planning, etc. (2) Realistic and controllable respiratory motion pattern. This feature is valuable for investigating 4D imaging sorting and reconstruction techniques, validating ventilation map calculations, functional image analysis, developing motion-robust treatment planning, etc. (3) Realistic onboard CBCT imaging artifacts. This feature is valuable for evaluating and optimizing the imaging protocol and various image processing and treatment techniques, such as CBCT artifact correction, CBCT based target localization, radiomics analysis, and plan adaptation.
Item Open Access Dose-Distribution-Driven PET Image-Based Outcome Prediction (DDD-PIOP): A Deep Learning Study for Oropharyngeal Cancer IMRT Application(Frontiers in Oncology) Wang, Chunhao; Liu, Chenyang; Chang, Yushi; Lafata, Kyle; Cui, Yunfeng; Zhang, Jiahan; Sheng, Yang; Mowery, Yvonne; Brizel, David; Yin, Fang-Fang