Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

We investigate the problem of automatic cardiomegaly diagnosis. We approach this by developing classifiers using multimodal data enhanced by two image-derived digital biomarkers, the cardiothoracic ratio (CTR) and the cardiopulmonary area ratio (CPAR). The CTR and CPAR values are estimated using segmentation and detection models. These are then integrated into a multimodal network trained simultaneously on chest radiographs and ICU data (vital sign values, laboratory values and metadata). We compare the predictive power of different data configurations with and without the digital biomarkers. There was a negligible performance difference between the XGBoost model containing only CTR and CPAR (accuracy 81.4%, F1 0.859, AUC 0.810) and black-box models which included full images (ResNet-50: accuracy 81.9%, F1 0.874, AUC 0.767; Multimodal: 81.9%, F1 0.873, AUC 0.768). We concluded that models incorporating domain knowledge-based digital biomarkers CTR and CPAR provide comparable performance to black-box multimodal approaches with the former providing better clinical explainability.

Original publication

DOI

10.1007/978-3-031-12053-4_2

Type

Conference paper

Publication Date

01/01/2022

Volume

13413 LNCS

Pages

13 - 27