KENDALL MCNEIL November 2023
DESCRIPTION: The chest radiograph is one of the most challenging to interpret, which can result in misdiagnosis even for seasoned healthcare providers. Building a strong convolutional neural network (CNN) to detect common thoracic lung diseases in chest x-rays would improve diagnostic accuracy for patients and ultimately save lives through early and accurate detection. The CNN will act as an automated system to support radiologists as a second opinion in reviewing chest x-rays for abnormalities. The work product will alleviate the stress of busy doctors and healthcare providers while also providing patients with a more accurate and efficient diagnosis.
OBJECTIVE: The objective, therefore, is to detect a variety (14 total) of common thoracic lung abnormalities in chest x-rays by building a Convolutional Neural Network (CNN) to develop an AI system for thoracic lung abnormality detection. The multi-label neural network model was designed using Tensorflow.
AUDIENCES: The general target audience for the project is healthcare providers. The more specific presentation audience is Vingroup Big Data Institute (VinBigData) that is working to build large-scale and high-precision medical imaging solutions based on the latest advancements in AI to facilitate efficient clinical workflows.
DATA: VinBigData has provided a dataset of 18,000 CXR scans dicom images labeled by a panel of experienced radiologists for the presence of 14 common thoracic abnormalities: aortic enlargement, atelectasis, calcification, cardiomegaly, consolidation, ILD, infiltration, lung opacity, nodule/mass, other lesion, pleural effusion, pleural thickening, pneumothorax, and pulmonary fibrosis. The dataset was created by assembling de-identified chest X-ray studies provided by two hospitals in Vietnam: the Hospital 108 and the Hanoi Medical University Hospital.