Abstract

1548 Background: The current lung cancer screening strategy has not had a major impact on public health, facing many challenges during its implementation in developing countries. Although chest radiography (CXR) is not the study of choice for screening; it is one of the most accessible studies used at all levels of care, so implementing an AI tool capable of CXR interpretation as well as lung cancer risk assessment on a large scale could improve the selection of patients for lung cancer screening. The aim of our study was to develop and evaluate this tool. Methods: The dataset used for this tool development was obtained from the NIH "ChestX-ray14” public database. The set was divided into 3 subgroups for training, testing, and validation in a ratio of 80/10/10% respectively. We performed a structuration of metadata and an exploratory analysis as well as preprocessing of the images to standardize their size and resolution before the training. A model based on the adaptation of the momentum for learning was used as an optimizer of the CNN algorithm used. Additionally, a questionnaire based on the PLCO clinical criteria was included for the identification of patients at high risk of developing lung cancer. Results: Standard performance metrics were evaluated as shown. Conclusions: Our model is comparable to certified radiologists in accuracy at identifying patterns of a CXR. The use of clinical patient information improves the pre-selection of patients with a high risk of developing lung cancer. In developing countries, the implementation of this tool on a large scale will allow more patients to be screened and early diagnosed.[Table: see text]

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call