Aleksandar (Alex) Vakanski


Research Interests

Research Projects

Machine Learning-based Analsyis of Breast Ultrasound Images

The project employs machine learning for computer-aided diagnosis of breast ultrasound images. It is a collaborative research project led by Dr. Min Xian from the Department of Computer Science at the University of Idaho. Our team developed approaches for integrating prior knowledge of visual saliency into deep learning-based tumor segmentation, design of explainable models for cancer diagnosis using the BI-RADS lexicon, detection of small-size tumors in breast images, and segmentation of histopathology images for breast cancer detection.


Human Movement Modeling in Biomedical Applications

The research aims at employing deep neural networks for automated evaluation of patient performance in physical rehabilitation. The modeling approach is based on a hierarchical multilayer architecture designed to handle spatial and temporal variations in captured movement data. The research also seeks to formulate criteria and metrics for quantifying patient performance in rehabilitation programs. Our team created the UI-PRMD dataset (University of Idaho – Physical Rehabilitation Movement Dataset) to facilitate the model training and validation.


Robot Programming by Demonstration

The goal of the project is commercializing a robotic system with abilities to learn new skills from visual observation of human demonstrated skill examples. The research expands on an approach for image-based robot programming by demonstration, which I developed during the Ph.D. studies. For this project, I co-founded a start-up company Visual Learning Robotics, which developed a prototype of the system. Task demonstrations are captured with a vision camera, and a learning robot employs machine learning for movement modeling and generates a strategy for task reproduction.


Crop Health Assessment in Precision Agriculture

The project focuses on image-processing for crop stress detection in aerial images. The specific aims of the project are: collect field images by using a multispectral camera carried by an unmanned aerial system; annotate the collected image data; and, design a neural network architecture for discrimination of healthy and diseased plants in the images. Early diagnosis of crop health symptoms can reduce the volume of chemical substances applied at later phases, and contribute to increased crop yield.