U-net BML Segmentation

Object Type: Asset
In Folder: U-net BML Segmentation

Expand to show object metadata

U-net BML Segmentation

Senior Design Showcase project, Wentworth Institute of Technology, College of Engineering and Computer Science, Department of Computer Science and Networking, 2020

Dublin Core Metadata

title: U-net BML Segmentation

creator: Garrett Stonis

subject: BML Machine learning U-net CNN Bone Osteoarthritis

description: BML (bone marrow lesions) is a clinical condition where fluid builds up in the bone marrow; it can also be an early sign of osteoarthritis. Early detection of BML could lead to the prevention of cartilage deterioration and an improved quality of life for the patient. Segmentation of BML (which is currently done manually) is important for doctors to diagnose the severity of the BML. A dataset of 300 knee MRI, with each BML manually segmented, was used in various experiments. Each experiment used U-net CNN as the main segmentation tool. First, the BML was attempted to be segmented from the entire image, which proved lackluster. Then, a bone segmentation model was created using manually segmented bone data. This model was used to crop the bone, then another model was trained to segment BML. It also failed because the bone model cut out the majority of BML in the image. Attempts to improve the pipeline by post-processing the bone model output via bounding box also proved unsuccessful. A final experiment used the manual bone data to crop the bone, then a BML model followed. This pipeline was promising. While the experiments failed to segment BML fully automatically, they proved that a reliable bone segmentation model can be used to more successfully segment BML via U-net models. Perhaps other pre/post-processing techniques would further improve the model.

publisher: The Douglas D. Schumann Library and Learning Commons, Wentworth Institute of Technology

contributor: Juan Shan, Yaodong Du, Rania Almajalid

date: 2020

type: Poster

format: .pdf

identifier:

source:

language: eng

relation:

coverage:

rights: https://rightsstatements.org/vocab/InC/1.0/

Show more ▼

U-net BML Segmentation

Other Documents

U-net BML Segmentation

U-net BML Segmentation

20.CSN.89Poster