Electrical Engineering and Systems Science > Image and Video Processing
[Submitted on 3 Nov 2023]
Title:CheX-Nomaly: Segmenting Lung Abnormalities from Chest Radiographs using Machine Learning
View PDFAbstract:The global challenge in chest radiograph X-ray (CXR) abnormalities often being misdiagnosed is primarily associated with perceptual errors, where healthcare providers struggle to accurately identify the location of abnormalities, rather than misclassification errors. We currently address this problem through disease-specific segmentation models. Unfortunately, these models cannot be released in the field due to their lack of generalizability across all thoracic diseases. A binary model tends to perform poorly when it encounters a disease that isn't represented in the dataset. We present CheX-nomaly: a binary localization U-net model that leverages transfer learning techniques with the incorporation of an innovative contrastive learning approach. Trained on the VinDr-CXR dataset, which encompasses 14 distinct diseases in addition to 'no finding' cases, my model achieves generalizability across these 14 diseases and others it has not seen before. We show that we can significantly improve the generalizability of an abnormality localization model by incorporating a contrastive learning method and dissociating the bounding boxes with its disease class. We also introduce a new loss technique to apply to enhance the U-nets performance on bounding box segmentation. By introducing CheX-nomaly, we offer a promising solution to enhance the precision of chest disease diagnosis, with a specific focus on reducing the significant number of perceptual errors in healthcare.
Current browse context:
cs
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.