Description:
Fast-Track proposals will not be accepted.
Number of anticipated awards: 2-3
Budget (total costs, per award):
Phase I: up to $225,000 for up to 9 months
Phase II: up to $1,500,000 for up to 2 years
PROPOSALS THAT EXCEED THE BUDGET OR PROJECT DURATION LISTED ABOVE MAY NOT BE FUNDED.
Summary
The goal of this topic is to see if Artificial Intelligence (AI) technology can be used to improve treatment planning for prostate cancer by developing algorithms to “read” standard Computerized Tomography (CT) images in context with clinical information and recommend suitable treatment plan approaches. The resulting tool may aid radiation oncologists in reaching unbiased consensus treatment planning, help train junior radiation oncologists, update practitioners, reduce professional costs, and improve quality assurance in clinical trials and patient care. Treatment planning for radiation therapy has become increasingly complex with the advent of image-guided radiation therapy and charged particle therapy. A substantial amount of physician time and effort are required to contour key tumor and normal tissue structures. The process involves assessing the patient’s risk for disease progression based on tumor volume; histological grade and biomarkers (e.g., prostate specific antigen or other tests); and assigning one of three risk groups as defined in the National Comprehensive Cancer Network (NCCN) guidelines: low, intermediate, or high. See NCCN guidelines here: https://www.trikobe.org/nccn/guideline/urological/english/prostate.pdf. Radiation treatment will use external beam radiation with or without androgen deprivation. Imaging uses CT and often magnetic resonance imaging. Based on these, the physician and medical physicists plan the target volume to be treated, radiation dose, and normal tissue to be spared. In practice, treatment guidelines are established by consensus papers. However, proposed plans among even world renowned experts often differ.
Thus, it may be possible to go beyond verbal consensus text and understand the rationale among expert “preferences” in
treatment plans by using AI-based contextual image analysis that uses feature extracting algorithms and/or interactive machine learning to formulate treatment plan. Such an approach would provide an initial plan to the physician upon which to facilitate treatment planning, build consensus, and help understand expert thinking.
Project Goals
The goal of this contract topic is to develop and evaluate the concept that AI can be used to understand and duplicate experts’ radiation therapy planning. The purpose is to understand how human cognition performs in work, focused in the context of developing radiation therapy treatment plans, and then incorporate such an understanding into machine learning with the intent to automate treatment planning to reduce subjective biases, improve treatment quality, and reduce cost. This contract topic does not intend to achieve a breakthrough in AI technology. The objective is to integrate recent advances in treatment planning systems and machine learning to improve radiation therapy by eliminating repetitive, time-consuming, and subjective biases in treatment delivery. Subjective biases could result in normal tissue injury and compromise therapeutic
benefit. Machine learning approaches may involve extraction of relevant features from “consensus image datasets” of expert
medical teams and then applying them to train machines with an initial focus on prostate cancer. The broad and highly impactful goal is to improve the outcome for patients with prostate cancer. By developing knowledge-based planning solutions, it may be possible to provide a more standardized treatment at a significantly lower cost. This may facilitate quality assurance, possibly extending it to facilities with limited expert personnel and enabling the conduct of research by reducing the variability and potential arbitrariness and/or preference that individuals incorporate in their treatment design. The goal of this project is to encourage creative small businesses to design, develop, and build approaches to AI-based treatment planning systems to improve radiation therapy. Progress here could be applied to other disease sites.
Activities supported by this topic:
Proposals that develop AI software that only outlines tumor and normal tissues but does not select a treatment plan for the three risk groups will not be considered for funding.
Phase I Activities and Deliverables
Establish a project team to develop an AI tool to understand and improve treatment planning for prostate cancer, comprising of cross disciplinary expertise. This cross disciplinary expertise will require proven expertise in AI, application development, radiation treatment planning for prostate cancer, IT experience in a healthcare setting, data security, HIPAA, and other laws and regulations to protect privacy and confidentiality of patient information.
o
Choose one expert radiation therapy planning team comprising of a physician and planners (i.e., a person who is knowledgeable in treatment planning with good understanding of the treatment planning system) and evaluate expert cognition process in developing treatment planning strategies for all three strata of patient risk groups (i.e., low, intermediate, and high based on NCCN guidelines).
o
Note: For Phase II, three independent teams will be required so that there can be comparison between expert teams using the same set of cases.
Identify criteria used by an expert planner to develop each treatment plan for each risk group (i.e., low, intermediate, and high NCCN).
o Based on the expert planning methods, develop an AI based planning process including feature extraction. Existing algorithms could be used for feature selection. Expected innovation is in using AI for treatment planning. Plan will use external beam radiation, fields to be determined by expert team, with or without androgen deprivation therapy per expert’s discretion. Other forms of treatment will not be considered for this project.
Design and develop computational algorithms/methods aimed at improving treatment planning for prostate cancer
patients.
Propose plan to develop, incorporate, and compare the AI methods with expert treatment planning methods and
validate AI based treatment planning system.
o There should be a minimum of 10 patients per risk group or a suitable number that the research team feels is sufficient for the AI algorithm to begin the initial planning. Provide justification for the selected number of patients. Retrospective de-identified data could be used for this purpose.
Present AI concept to develop knowledge based radiotherapy treatment planning to NCI’s SBIR Development Center
and the Radiation Research Program.
Design and deliver an AI approach to improve radiation therapy planning for prostate cancer to be tested in Phase II.
Present an estimate of the number of training and validation sets that would be needed for each of the risk groups so that the AI results can provide a starting point for the planning team to refine the initial plan and determine the final course of treatment.
o Establish a set of patient records for the three risk groups to be used in Phase II among the 3 expert teams.
At a minimum, apply this technology to standard 3D CT datasets. Use of additional imaging is at the preference of the planning team.
Phase II Activities and Deliverables
Refinement of algorithm based on the results of Phase I.
Demonstration of utility of AI plan as the initial step to be reviewed and then modified by the planning team.
Establish sufficient cases in each of the three treatment categories for the comparison among expert groups, based on Phase I deliverables.
Expand to a minimum of three independent expert treatment planning teams and have each expert team plan the 3 risk
groups. There will be 3 consensus reviews: a) comparing the plans among the 3 expert panels done by the “standard”
hands-on approach; b) comparison of the 3 AI produced plans; and, c) an analysis of how the hands-on and AI based
plans differ.
Compare the consensus approach of the expert hands-on plans to the consensus among the AI plans to see where there was agreement or disagreement and see if this difference can be understood and rectified. This would enable the AI to refine its algorithm.
Evaluate developed AI software to see if it can match the performance of the expert teams (each team would have 3 categories of patients). Examine differences and present plans to refine the performance of the AI.
Expand types of data sets to include MRI or PET or other sources of data that would improve AI’s performance. Establish external partnership(s) for future validation of method, as demonstrated with letters of intent from strategic partners.
