Deep Learning of CT Virtual Wedge Resection for Prediction of Histologic Usual Interstitial Pneumonitis.

Researchers

Journal

Modalities

Models

Abstract

The CT pattern of definite or probable usual interstitial pneumonia (UIP) can be diagnostic of idiopathic pulmonary fibrosis (IPF) and may obviate the need for invasive surgical biopsy. Few machine learning studies have investigated the classification of interstitial lung disease (ILD) on CT, but none have used histopathology as a reference standard.
Predict histopathologic UIP using deep learning of high-resolution CT (HRCT).
Institutional databases were retrospectively searched for consecutive patients with ILD, HRCT and diagnostic histopathology from 2011-2014 (training cohort) and from 2016-2017 (testing cohort). A blinded expert radiologist and pulmonologist reviewed all training HRCTs in consensus and classified HRCT scans based on the 2018 ATS/ERS/JRS/ALAT diagnostic criteria for IPF. A convolutional neural network (CNN) was built accepting 4x4x2 cm virtual wedges of peripheral lung on HRCT as input and outputting the UIP histopathologic pattern. The CNN was trained and evaluated on the training cohort utilizing 5-fold cross validation and then tested on the hold out testing cohort. CNN and human performance were compared in the training cohort. Logistic regression and survival analyses were performed.
The CNN was trained on 221 patients (median age 60 years, IQR 53-66), including 71 (32%) with “UIP” or “probable UIP” histopathologic patterns. The CNN was tested on a separate hold out cohort of 80 patients (median age 66 years, IQR 58-69), including 22 (27%) with “UIP” or “probable UIP” histopathologic patterns. An average of 516 wedges were generated per patient. Percent of wedges with CNN-predicted UIP yielded a cross-validation AUC of 74% for histopathological UIP pattern per patient. 16.5% of virtual lung wedges with CNN-predicted UIP was the optimal cutoff point for classifying patients on the training cohort and resulted in sensitivity and specificity of 74% and 58% in the testing cohort, respectively. CNN-predicted UIP was associated with an increased risk of death or lung transplantation during cross validation (hazard ratio 1.5, 95% CI 1.1-2.2, p-value=0.03).
Virtual lung wedge resection in patients with ILD can be used as input to a CNN for predicting the histopathologic UIP pattern and transplant-free survival.