Senior Design Projects

ECS193 A/B Winter & Spring 2021

Web-based Liver Image Segmentation based on Deep Learning

Email **********
Name Roger Goldman, MD, PhD
Affiliation UC Davis School of Medicine, Department of Radiology

Project's details

Project title Web-based Liver Image Segmentation based on Deep Learning
Background Liver cancer is the sixth most common cancer diagnosis and the fourth most frequent cause of cancer-related death globally. In the United States in 2020, an estimated 42,810 new cases of liver cancer will be diagnosed with a mortality estimate of 30,1602. Globally and within the United States, liver cancer is one of only two cancers to have an increasing mortality rate year-over-year over the past 20 years. Understanding and clearly defining the anatomic structure of the liver, including the vascular structures and fluid conduits that traverse the parenchyma, are critical to effective staging and treatment of liver cancer. The anatomic structures may be delineated on cross-sectional imaging studies including CT and MRI. The current clinical process across the majority of academic and community medical centers is manual segmentation of medical images. This laborious process fails to leverage the extensive literature investigating automatic segmentation algorithms using deep learning models. Furthermore, no practical systems are available for evaluation and facile manual revision of automatically generated segmentation data. The result of this Senior Design Project will be unique system with the potential to facilitate treatment of liver cancer patients throughout the UC system.
Description The student team will develop a cloud-based service accessible via a web browser for optimal liver segmentation based upon a combination of automatic and manual techniques. The project will involve two separate development subtasks that can be performed in parallel. Subtask 1 will be development of a web-browser based service for viewing and segmenting DICOM image data. Subtask 2 will be training of a deep learning model for automatic liver segmentation based upon a robust dataset of labeled clinical images. The subtasks may be integrated to produce an end-to-end solution for robust and facile liver segmentation.
Deliverable 1) Development of a fast and intuitive web service to: a. accept DICOM images from a remote server b. accept image segmentation masks from a remote server c. provide an interface for image viewing d. provide tools to manipulate and save segmentation masks. 2) Development a cloud-based automatic liver segmentation algorithm by: a. selection or development a deep learning model for liver segmentation b. training the model using labeled liver segmentation dataset c. development of a pipeline to accept medical image data and return segmentation mask 3) System integration of the image web service with the automatic segmentation algorithm 4) Clear and thoroughly documented source code maintained in a GitHub repository
Skill set desirable The ideal team will be highly motivated to work with medical imaging data and clinicians. Proficiency with Python, web development (HTML, CSS, Javascript), and Cloud Platforms (Google Cloud and/or Amazon AWS) is highly desirable. Prior experience coding in TensorFlow, PyTorch, or other deep learning framework is also desired.
Phone number **********
Client time availability 30-60 min weekly or more
IP requirement Open source project
Attachment Click here
Selected Yes
Team members N/A
TA Albara