Analysis of Validation Performance of a Machine Learning Classifier in Interstitial Lung Disease Cases Without Definite or Probable Usual Interstitial Pneumonia Pattern on CT Using Clinical and Pathology-Supported Diagnostic Labels

• DOI: 10.1007/s10278-023-00914-w
• Publication Date: 2024-01-11T17:02:19Z
• Authors: Marcello Chang; Joshua J. Reicher; Angad Kalra; Michael Muelly; Yousef Ahmad
• Keywords: Usual interstitial pneumonia; Honeycombing

We previously validated Fibresolve, a machine learning classifier system that non-invasively predicts idiopathic pulmonary fibrosis (IPF) diagnosis. The incorporates an automated deep algorithm analyzes chest computed tomography (CT) imaging to assess for features associated with fibrosis. Here, we performance in assessment of patterns beyond those are characteristic usual interstitial pneumonia (UIP) pattern. was developed and using standard training, validation, test sets, clinical plus pathologically determined ground truth. multi-site 295-patient validation dataset used focused subgroup analysis this investigation evaluate the classifier's range cases without radiologic UIP probable designations. Radiologic specific including presence distribution reticulation, glass, bronchiectasis, honeycombing assignment Output from assessed within various subgroups. able classify not meeting criteria or as IPF estimated sensitivity 56–65% specificity 92–94%. Example demonstrated non-basilar-predominant well glass were indeterminate by subjective but which correctly identify case confirmed multidisciplinary discussion generally inclusive histopathology. Fibresolve may be helpful diagnosis radiological patterns.