Daniel Polan

Tissue segmentation of computed tomography images using a Random Forest algorithm: a feasibility study

There is a need for robust, fully automated whole body organ segmentation for diagnostic CT. This study investigates and optimizes a Random Forest algorithm for automated organ segmentation; explores the limitations of a Random Forest algorithm applied to the CT environment; and demonstrates segmentation accuracy in a feasibility study of pediatric and adult patients. To the best of our knowledge, this is the first study to investigate a trainable Weka segmentation (TWS) implementation using Random Forest machine-learning as a means to develop a fully automated tissue segmentation tool developed specifically for pediatric and adult examinations in a diagnostic CT environment. Current innovation in computed tomography (CT) is focused on radiomics, patient-specific radiation dose calculation, and image quality improvement using iterative reconstruction, all of which require specific knowledge of tissue and organ systems within a CT image. The purpose of this study was to develop a fully automated Random Forest classifier algorithm for segmentation of neck-chest-abdomen-pelvis CT examinations based on pediatric and adult CT protocols. Seven materials were classified: background, lung/internal air or gas, fat, muscle, solid organ parenchyma, blood/contrast enhanced fluid, and bone tissue using Matlab and the TWS plugin of FIJI. The following classifier feature filters of TWS were investigated: minimum, maximum, mean, and variance evaluated over a voxel radius of 2 (n) , (n from 0 to 4), along with noise reduction and edge preserving filters: Gaussian, bilateral, Kuwahara, and anisotropic diffusion. The Random Forest algorithm used 200 trees with 2 features randomly selected per node. The optimized auto-segmentation algorithm resulted in 16 image features including features derived from maximum, mean, variance Gaussian and Kuwahara filters. Dice similarity coefficient (DSC) calculations between manually segmented and Random Forest algorithm segmented images from 21 patient image sections, were analyzed. The automated algorithm produced segmentation of seven material classes with a median DSC of 0.86  ±  0.03 for pediatric patient protocols, and 0.85  ±  0.04 for adult patient protocols. Additionally, 100 randomly selected patient examinations were segmented and analyzed, and a mean sensitivity of 0.91 (range: 0.82-0.98), specificity of 0.89 (range: 0.70-0.98), and accuracy of 0.90 (range: 0.76-0.98) were demonstrated. In this study, we demonstrate that this fully automated segmentation tool was able to produce fast and accurate segmentation of the neck and trunk of the body over a wide range of patient habitus and scan parameters.

SU-F-J-85: Evaluation of the Velocity Deformable Image Registration Algorithm

PURPOSE To perform validation and commissioning of a commercial deformable image registration (DIR) algorithm (Velocity, Varian Medical Systems) for numerous clinical sites using single and multi-modality images. METHODS In this retrospective study, the DIR algorithm was evaluated for 10 patients in each of the following body sites: head and neck (HN), prostate, liver, and gynecological (GYN). HN DIRs were evaluated from planning (p)CT to re-pCT and pCTs to daily CBCTs using dice similarity coefficients (DSC) of corresponding anatomical structures. Prostate DIRs were evaluated from pCT to CBCTs using DSC and target registration error (TRE) of implanted RF beacons within the prostate. Liver DIRs were evaluated from pMR to pCT using DSC and TRE of vessel bifurcations. GYN DIRs were evaluated between fractionated brachytherapy MRIs using DSC of corresponding anatomical structures. RESULTS Analysis to date has given average DSCs for HN pCT-to-(re)pCT DIR for the brainstem, cochleas, constrictors, spinal canal, cord, esophagus, larynx, parotids, and submandibular glands as 0.88, 0.65, 0.67, 0.91, 0.77, 0.69, 0.77, 0.87, and 0.71, respectively. Average DSCs for HN pCT-to-CBCT DIR for the constrictors, spinal canal, esophagus, larynx, parotids, and submandibular glands were 0.64, 0.90, 0.62, 0.82, 0.75, and 0.69, respectively. For prostate pCT-to-CBCT DIR the DSC for the bladder, femoral heads, prostate, and rectum were 0.71, 0.82, 0.69, and 0.61, respectively. Average TRE using implanted beacons was 3.35 mm. For liver pCT-to-pMR, the average liver DSC was 0.94 and TRE was 5.26 mm. For GYN MR-to-MR DIR the DSC for the bladder, sigmoid colon, GTV, and rectum were 0.79, 0.58, 0.67, and 0.76, respectively. CONCLUSION The Velocity DIR algorithm has been evaluated over a number of anatomical sites. This work functions to document the uncertainties in the DIR in the commissioning process so that these can be accounted for in the development of downstream clinical processes. This work was supported in part by a co-development agreement with Varian Medical Systems.

SU-F-J-89: Assessment of Delivered Dose in Understanding HCC Tumor Progression Following SBRT.

PURPOSE It is well documented that the delivered dose to patients undergoing radiotherapy (RT) is often different from the planned dose due to geometric variability and uncertainties in patient positioning. Recent work suggests that accumulated dose to the GTV is a better predictor of progression compared to the minimum planned dose to the PTV. The purpose of this study is to evaluate if deviations from the planned dose can contributed to tumor progression. METHODS From 2010 to 2014 an in-house Phase II clinical trial of adaptive stereotactic body RT was completed. Of the 90 patients enrolled, 7 patients had a local recurrence defined on contrast enhanced CT or MR imaging 3-21 months after completion of RT. Retrospective dose accumulation was performed using a biomechanical model-based deformable image registration algorithm (DIR) to accumulate the dose based on the kV CBCT acquired prior to each fraction for soft tissue alignment of the patient. The DIR algorithm was previously validated for geometric accuracy in the liver (target registration error = 2.0 mm) and dose accumulation in a homogeneous image, similar to a liver CBCT (gamma index = 91%). Following dose accumulation, the minimum dose to 0.5 cc of the GTV was compared between the planned and accumulated dose. Work is ongoing to evaluate the tumor control probability based on the planned and accumulated dose. RESULTS DIR and dose accumulation was performed on all fractions for 6 patients with local recurrence. The difference in minimum dose to 0.5 cc of the GTV ranged from -0.3-2.3 Gy over 3-5 fractions. One patient had a potentially significant difference in minimum dose of 2.3 Gy. CONCLUSION Dose accumulation can reveal tumor underdosage, improving our ability to understand recurrence and tumor progression patterns, and could aid in adaptive re-planning during therapy to correct for this. This work was supported in part by NIH P01CA059827.

SU-C-207B-05: Tissue Segmentation of Computed Tomography Images Using a Random Forest Algorithm: A Feasibility Study

PURPOSE Develop an automated Random Forest algorithm for tissue segmentation of CT examinations. METHODS Seven materials were classified for segmentation: background, lung/internal gas, fat, muscle, solid organ parenchyma, blood/contrast, and bone using Matlab and the Trainable Weka Segmentation (TWS) plugin of FIJI. The following classifier feature filters of TWS were investigated: minimum, maximum, mean, and variance each evaluated over a pixel radius of 2n, (n = 0-4). Also noise reduction and edge preserving filters, Gaussian, bilateral, Kuwahara, and anisotropic diffusion, were evaluated. The algorithm used 200 trees with 2 features per node. A training data set was established using an anonymized patients (male, 20 yr, 72 kg) chest-abdomen-pelvis CT examination. To establish segmentation ground truth, the training data were manually segmented using Eclipse planning software, and an intra-observer reproducibility test was conducted. Six additional patient data sets were segmented based on classifier data generated from the training data. Accuracy of segmentation was determined by calculating the Dice similarity coefficient (DSC) between manual and auto segmented images. RESULTS The optimized autosegmentation algorithm resulted in 16 features calculated using maximum, mean, variance, and Gaussian blur filters with kernel radii of 1, 2, and 4 pixels, in addition to the original CT number, and Kuwahara filter (linear kernel of 19 pixels). Ground truth had a DSC of 0.94 (range: 0.90-0.99) for adult and 0.92 (range: 0.85-0.99) for pediatric data sets across all seven segmentation classes. The automated algorithm produced segmentation with an average DSC of 0.85 ± 0.04 (range: 0.81-1.00) for the adult patients, and 0.86 ± 0.03 (range: 0.80-0.99) for the pediatric patients. CONCLUSION The TWS Random Forest auto-segmentation algorithm was optimized for CT environment, and able to segment seven material classes over a range of body habitus and CT protocol parameters with an average DSC of 0.86 ± 0.04 (range: 0.80-0.99).

WE-AB-BRA-02: Development of Biomechanical Models to Describe Dose-Volume Response to Liver Stereotactic Body Radiation Therapy (SBRT) Patients

Purpose: Previous studies have shown that radiotherapy treatment for liver metastases causes marked liver hypertrophy in areas receiving low dose and atrophy/fibrosis in areas receiving high dose. The purpose of this work is to develop and evaluate a biomechanical model-based dose-response model to describe these liver responses to SBRT. Methods: In this retrospective study, a biomechanical model-based deformable registration algorithm, Morfeus, was expanded to include dose-based boundary conditions. Liver and tumor volumes were contoured on the planning images and CT/MR images three months post-RT and converted to finite element models. A thermal expansion-based relationship correlating the delivered dose and volume response was generated from 22 patients previously treated. This coefficient, combined with the planned dose, was applied as an additional boundary condition to describe the volumetric response of the liver of an additional cohort of metastatic liver patients treated with SBRT. The accuracy of the model was evaluated based on overall volumetric liver comparisons and the target registration error (TRE) using the average deviations in positions of identified vascular bifurcations on each set of registered images, with a target accuracy of the 2.5mm isotropic dose grid (vector dimension 4.3mm). Results: The thermal expansion coefficient models the volumetric change of the liver to within 3%. The accuracy of Morfeus with dose-expansion boundary conditions a TRE of 5.7±2.8mm compared to 11.2±3.7mm using rigid registration and 8.9±0.28mm using Morfeus with only spatial boundary conditions. Conclusion: A biomechanical model has been developed to describe the volumetric and spatial response of the liver to SBRT. This work will enable the improvement of correlating functional imaging with delivered dose, the mapping of the delivered dose from one treatment onto the planning images for a subsequent treatment, and will further provide information to assist with the biological characterization of patients’ response to radiation.