Photoacoustic Imaging for Efficient Dataset Automation in Deep Learning Aided Ultrasound-Guided Needle Tracking
Date
2024-12
Authors
Gisi, Katherine
Major Professor
Pramanik, Manojit
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Deep learning (DL) for biomedical imaging is a dynamic research field, encompassing countless clinical practices from diagnostic imaging to procedural monitoring. However, there remain significant systematic challenges including the tediousness of dataset annotation. This is found especially in biomedical imaging where annotation is often based on manual expert review. One of these DL-aided imaging practices is ultrasound (US)-guided needle-tracking. US-guided needle tracking is a minimally invasive technique used for peripheral nerve blocks, tumor biopsies, fetal blood sampling etc. However, the visualization of the needle under US guidance require expert sonographer skills and training. Therefore, a DL-enhanced needle visualization during US-guided needle tracking could increase the range of biological and medical applications and ease of use of the needle tracking. In this work we have used photoacoustic (PA) image as a ground truth to train the DL network, which later predicts the needle location automatically in the ultrasound images. First, we have created a dataset preparation pipeline. This automates the process of transforming recorded US +PA images to a robust training dataset. A 24 second US/PAI data acquisition produces up to 1000 images and their corresponding masks. A LED-based handheld PA imaging system (Cyberdyne) was used for the data collection. The separation and cleaning of the data is done through various image processing blocks. Preliminary results from networks, using the automatically annotated datasets, exhibit sufficient accuracy for segmentation of needles. Developing a dataset creation pipeline can remove the need for resource-consuming annotation and system characterization.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
creative component
Comments
Rights Statement
Attribution-NonCommercial-NoDerivs 3.0 United States
Copyright
2024