Mammo-Find: An LLM-based Visual App for Matching Public Mammogram Datasets to Clinical and Translational Pipelines

Click to View Presentation

Presentation Time: 10:45 AM - 11:00 AM

Abstract Keywords: Data Visualization, Artificial Intelligence/Machine Learning, Data Science
Primary Track: AI and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

For assisting the physicians, the use of AI has increased manyfold to diagnose and treat breast cancer through mammogram images. AI models need a lot of data to carry out such sensitive tasks. However, mammogram data are not readily available. So, researchers need to spend considerable time looking for suitable datasets. Our developed tool Mammo-Find, can help such researchers to find out suitable dataset for their work quickly in the form of a visualization tool.

Speaker:
Raiyan Jahangir, PhD Student
University of California, Davis

Authors:
Raiyan Jahangir, PhD Student - University of California, Davis; Vladimir Filkov, PhD - University of California, Davis;

Topological feature extraction from 3D images for Alzheimer’s disease classification

Click to View Presentation

Presentation Time: 11:00 AM - 11:15 AM

Abstract Keywords: Artificial Intelligence/Machine Learning, Data Science, Big Data, Patient-Generated Data / Patient Reported Outcomes (PROs)
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

In this study, we propose a novel feature extraction method using persistent homology to analyze structural MRI of the brain. This approach converts topological features into powerful feature vectors through Betti functions. By integrating these feature vectors with a simple machine learning model like XGBoost, we achieve a computationally efficient machine learning (ML) model. Our model outperforms state-of-the-art deep learning models in both binary and three-class classification tasks for ADNI 3D MRI disease diagnosis. Using 10-fold cross-validation, our model achieved an average accuracy of 97.43% and sensitivity of 99.09% for binary classification. For three-class classification, it achieved an average accuracy of 95.47% and sensitivity of 94.98%. Unlike many deep learning models, our approach does not require data augmentation or extensive preprocessing, making it particularly suitable for smaller datasets. Topological features differ significantly from those commonly extracted using convolutional filters and other deep learning machinery. Because it provides an entirely different type of information from ML models, it has the potential to combine topological features with other ML models later on.

Speaker:
Mohammad Alfrad Nobel Bhuiyan, PhD
LSU Health Shreveport

Machine learning model to identify fast-track patients in the emergency department waiting room.

Click to View Presentation

2025 Clinical Informatics Conference DEI/Health Equity Presentation

Presentation Time: 11:15 AM - 11:30 AM

Abstract Keywords: Workflow Efficiency, Patient Safety, Artificial Intelligence/Machine Learning, Cloud Computing and Storage
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Many emergency departments (ED) implement a “fast-track” system that is designed to identify patients in the waiting room who are likely to have short length of stay (LOS) and room them in a dedicated area where they are quickly dispositioned. Identification of such patients is based on provider gestalt and could be inaccurate. Existing artificial intelligence (AI) models are designed to predict mean or average LOS, rather than focusing on LOS prediction for individual patients. Our study is aimed to develop a machine learning (ML) model to predict which specific patients in the ED waiting room are likely to have ED room to disposition time under 60 minutes and therefore would qualify for “fast-track” rooming.

We included all adult ED visits at our academic institution from 01/01/2022 to 12/31/2023. We created an ML model that used 80% of the dataset for training and 20% for testing. Input variables included demographics and triage data, means of arrival, chronic diseases, recursive data, and orders placed while in the waiting room. 

A total of 94,572 unique ED visits were included in the analysis. The model demonstrated the following performance for predicting which patients were likely to have room to disposition time under 60 minutes: area under the curve (AUC) 0.84, accuracy 0.90, precision 0.53, and recall 0.18.

The variables that correlate with lower LOS are similar to prior research. However, this model is designed to assist clinicians by identifying specific patients that should be considered for “fast-track” placement.

Speaker:
Daria Hunter, MD
Mayo Clinic

Authors:
Brendan Carr, MD, MBA - Mayo Clinic; Jacob Morey, MD, MBA - Mayo Clinic; Derick Jones, MD/MBA - Mayo Clinic;