NYC Gazette

Dr. Youshan Zhang, an assistant professor at the Katz School of Science and Health, has secured a $175,000 grant from the National Science Foundation (NSF). The funding is for research focused on improving cardiomegaly diagnosis in animals through the development of an AI-driven diagnostic tool.

Cardiomegaly, or heart enlargement, is a significant early indicator of heart disease, particularly in dogs. It remains one of the leading causes of death in both humans and animals. Traditionally, this condition is detected through manual analysis of thoracic radiographs using the Vertebral Heart Scale (VHS), a method that can be time-consuming and prone to human error.

Dr. Zhang’s project titled "Cardiac Disease Detection with AI for Veterinary Medicine" seeks to address these challenges by developing deep learning models that automate the VHS process with greater accuracy and speed. Dr. Zhang stated, “The primary goal of the project is to bridge the gap between traditional clinical methods and advanced AI models.” He noted that many clinicians find it difficult to trust AI-generated results due to a lack of transparency in current models.

To build trust and usability, Dr. Zhang's research aims to integrate traditional VHS metrics into a deep learning framework. This integration will help clinicians understand how AI-derived predictions align with established medical standards. The focus will be on enhancing model transparency and accuracy to make diagnoses more intuitive for veterinary professionals.

This project builds upon Dr. Zhang’s previous work published in Scientific Reports where he introduced the Regressive Vision Transformer (RVT) for assessing cardiomegaly in dogs. The new initiative outlines three main goals:

1. Development of New Detection Models: A new tool called a perpendicular fully connected layer (PFCL) will be created to improve measurement precision in X-ray images by ensuring right angles in measurements.

2. Automatic Report Generation: Tools capable of generating cardiomegaly reports with minimal training data will be developed using deep semantic mapping techniques.

3. User-Friendly Software Interface: An accessible software interface combining data labeling, result prediction, report generation, and modification will be created for easy use by clinicians without requiring prior domain knowledge.

Dr. Zhang emphasized the broader impact of this work by saying, “By developing a more precise and accessible diagnostic tool, the project aims to lower the cost of cardiomegaly detection while improving diagnostic accuracy and reducing stress for pet owners.” He also suggested that these deep learning models could inspire similar applications in human medicine for early heart disease detection.