TissueDetection2024

Abstract

Medical ultrasounds visualize internal structures, like organs. A physician interpret- ing an ultrasound image must identify structure contours, brightness, and texture differences. YOLOv8 is a state of the art image segmentation model that indicates the contours of objects in an image. This study develops a real-time segmentation model that identifies tissue in ultrasound images. Ultrasound images were obtained, spliced into frames, and annotated in Roboflow. The Ultralytics library and Com- mand Line Interface trained an image segmentation model. After 111 epochs of training, the model’s performance increased. The 86th epoch produced the best model. It had a 0.85115 mAP50-95 rating and 0.91 F1 Score at 0.521 confidence. The model precisely identifies tissue in a 60 test image assessment. The model reliably predicted images in 300 ms or less. This study developed a model that can be utilized in real-time clinical practice. One application is automated demarcation of tissue with three dimensional modeling scans. Future studies integrating in-vivo organ structure ultrasound scans into the training dataset would develop a model competent for clinical practice.

Greetings

Hello! My name is Zain Shariff, I am a freshman at Curtis Junior High School, and I am deeply excited to showcase my research that I have been working for a year. This year's project consists of the development of an image machine learning model that detects tissue present in ultrasound scans, and detects and trace contours of that tissue.

How did you do this?

To obtain the images, I used a 3.5 MHz ultrasound as parsed it using a simple python program I had written on my local system. I imported the images onto Roboflow which were annotated. I exported the dataset and developed a YOLOv8 extra large segmentation model using Ultralytics. This was done on a local system using Command Line Interface CLI. I have gone more in-depth into the process in my paper.

Where do I find what?

To find the images used for the dataset go into the dataset folder. The Roboflow Dataset can be found here. For the parser code in python go to the the python file named "splicer.py". The model can be found as "best.pt". My paper can be found in "paper.pdf".

License

Tissue Detection in Ultrasound Images Using Real-Time Image Segmentation Model © 2024 by Zain Shariff is licensed under CC BY-NC-SA 4.0