In 2015, nearly 1.7 million new cancer cases are estimated to occur in the US, where nearly two thirds will have treatment by radiation therapy. In current clinical practice, organ contour delineation on medical images is still performed with low levels of automation due to lack of highly automated commercial contouring software. This deteriorates therapy planning due to several reasons. (1) Manual contouring is error-prone, subject to intra- and inter-observer variations, labor intensive, and results in suboptimal throughput. (2) In modern conformal radiation techniques like intensity modulated radiotherapy and proton beam radiation therapy (PBRT), anatomic changes taking place during a 5-8 week course of serial treatment are not accounted for due to manual labor involved in re-contouring. (3) Such changes can significantly affect the total dose delivered to the tumor and normal surrounding organs and are particularly important when treating thoracic malignancies that are often highly radiosensitive. Given the above gaps and the fact that there are over 2,100 radiation therapy centers in the US, there is a strong commercial opportunity for reducing the total dose to normal critical structures and delivering highly precise radiotherapy to tumors, that can lead to improved patient outcomes, throughput, and cost-saving. The overall aim of this project, therefore, is to develop a methodology and a software prototype for clinical use to routinely define, at a high level of automation and improved accuracy, anatomical organ contours on CT and PET/CT images, and to evaluate the clinical utility of the software in thoracic radiation therapy.
The specific aims of this small business effort are 2-fold:
Aim 1 : To develop a highly automated method and software prototype to delineate all major thoracic organs on An automatic anatomy recognition methodology will be developed by adapting the highly successful fuzzy anatomy modeling technology developed at Penn to existing image and contour data from 200 lung cancer patients. Patients will be divided into four groups by gender and age, and modeling will be done for each group.
Aim 1 outcome will be a prototype software technically validated to have a mean false positive and false negative volume fractions at or below 5% (or Dice coefficient at or above 95%) and boundary distance within one voxel, as compared to reference segmentations. The expected human time needed in contouring will be 3 minutes or less per study.
Aim 2 : To perform a preliminary assessment of the utility of the proposed software for optimized radiation therapy planning in patients with thoracic malignancies treated with PBRT. Image and contour data of an additional 30 patients who have received proton treatment for thoracic malignancy and have undergone weekly CT studies in their routine clinical care will be studied, and the accuracy and efficiency of AAR software will b evaluated. Expected Aim 2 outcome will be similar to that described in Aim 1 but for the above patient cohort undergoing serial treatment. diagnostic CT images and low-dose CT of PET/CT acquisitions.
In 2015, 1.7 M new cancer cases are estimated to occur in the US, where nearly two thirds will have treatment that may involve radiation therapy. In current clinical practice, organ contour delineation on medical images by using commercial radiotherapy planning systems is still performed with low levels of automation due to lack of highly automated contouring software. This deteriorates therapy planning efficiency, throughput, and accuracy. This project aims to develop a methodology and a software prototype for clinical use to routinely define, at a high level of automation and improved accuracy, anatomical organ contours on CT and PET/CT images, and to evaluate the clinical utility of the software in thoracic radiation therapy. There is a strong commercial opportunity for such a software which can impart considerable impact on current practice of radiotherapy planning.
| Xu, Guoping; Udupa, Jayaram K; Tong, Yubing et al. (2018) Thoracic lymph node station recognition on CT images based on automatic anatomy recognition with an optimal parent strategy. Proc SPIE Int Soc Opt Eng 10574: |
| Yan, Fengxia; Udupa, Jayaram K; Tong, Yubing et al. (2018) Automatic Anatomy Recognition using Neural Network Learning of Object Relationships via Virtual Landmarks. Proc SPIE Int Soc Opt Eng 10574: |
| Wu, Xingyu; Udupa, Jayaram K; Tong, Yubing et al. (2018) Auto-contouring via Automatic Anatomy Recognition of Organs at Risk in Head and Neck Cancer on CT images. Proc SPIE Int Soc Opt Eng 10576: |
| Tong, Yubing; Udupa, Jayaram K; Wu, Xingyu et al. (2018) Hierarchical model-based object localization for auto-contouring in head and neck radiation therapy planning. Proc SPIE Int Soc Opt Eng 10578: |
| Pednekar, Gargi V; Udupa, Jayaram K; McLaughlin, David J et al. (2018) Image Quality and Segmentation. Proc SPIE Int Soc Opt Eng 10576: |
| Tong, Yubing; Udupa, Jayaram K; Odhner, Dewey et al. (2017) Virtual Landmarks. Proc SPIE Int Soc Opt Eng 10135: |
| Bai, PeiRui; Udupa, Jayaram K; Tong, YuBing et al. (2017) Automatic thoracic body region localization. Proc SPIE Int Soc Opt Eng 10134: |
