PFI-TT: Trustworthy Artificial Intelligence for the Volumetric Evaluation of Brain Tumors

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project consists of an innovative technology for developing trustworthy artificial intelligence (AI) models with applications in brain cancer monitoring and treatment. If successful, this technology will enable doctors who treat brain cancers to quickly identify and mark the boundaries of tumors in brain magnetic resonance imaging (MRI) scans. The accurate and reliable marking of boundaries of brain tumors is essential for many reasons, including planning future treatments, quantifying the effect of current treatment, or monitoring the progression of the cancerous tumor over time. The trustworthy AI models that the team is developing are expected to be applicable to other medical imaging indications. In addition, these models could be easily extended to other engineering areas that require accurate, highly reliable, and robust technological solutions. Given the current growth in the applications of AI models, the challenges of the trustworthiness and reliability of AI models will increase significantly. The technology being developed under this award may address these challenges and positively impact the lives of millions of people through societal applications in areas such as medicine, intelligent transportation systems, and manufacturing. The proposed project will develop a trustworthy AI-based brain tumor segmentation solution to identify and mark boundaries of tumors in brain MRI scans. The model is called the medical image Segmentation with Uncertainty Propagation in Encoder-Decoder Networks (SUPER-Net). A defining characteristic of SUPER-Net is that it simultaneously outputs a pixel-level tumor segmentation map and a corresponding pixel-level uncertainty map. The segmentation map identifies the tumor’s boundaries in an MRI, and the uncertainty map conveys the model’s confidence in its segmentations. This is achieved by propagating belief or uncertainty in the data and the model’s parameters during the model’s training. The project will focus on developing and validating SUPER-Net architecture for segmenting tumors in brain MRIs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.