Improving Signal Searches Under High-Noise Conditions

Signal searches under high background are ubiquitous challenges across numerous scientific fields, including medical imaging, weather forecasting, and astrophysical explorations. Here, the term background denotes noise sources that can mask the desired signal. Mis-specifying the background can lead to false detection or missed signals, thus compromising the accuracy and reliability of the detection.

PhD student Xiangyu Zhang (Statistics), in a project called “Searching for new signals under high background: A novel inferential framework,” is extending previous work that developed a unifying framework to solve this problem on both discrete and continuous independent and identically distributed (IID) data. In many practical applications, the data collection tools adopted (e.g., cameras, telescopes, etc.) often lead to binned or grouped data. In these settings, the data consists of event counts observed on different cells and the data can no longer be identically distributed. Therefore, extending the previous framework to incorporate the binned data regime is needed. This methodological development would enable a variety of different applications. For example, it can become an indispensable tool for the analysisof images generated by single-photon emission computed tomography (SPECT) in investigating blood flow, metabolic processes, and other variables that may contribute to the detection and identification of tumor markers or other diseases such as Parkinson’s disease and stroke.

Some funding for this project was provided by a DSI-MnDRIVE PhD Graduate Assistantship. The DSI-MnDRIVE Graduate Assistantship program supports U of M PhD candidates pursuing research at the intersection of data science and any of the five MnDRIVE areas: 

  • Robotics
  • Global Food 
  • Environment
  • Conditions
  • Cancer Clinical Trials

Projects in this program must align with one of the Data Science Tracks:

  • Foundational Data Science
  • Digital Health and Personalized Health Care Delivery

This project is part of the Environment, Brain Conditions, and Cancer Clinical Trials MnDRIVE areas and the Foundational Data Sciences track. See the complete list of the RC-MnDRIVE Graduate Assistantships.

Image description: The project goal is to compare the spectrum of RT Cru with a theoretical model. The counts spectrum is shown as the black histogram and the sum of the estimated background and the predicted model is shown in red.

A black historgram shows the counts spectrum and a red line shows the sum of the estimated background and the predicted model.