Improving prognostic accuracy in lung transplantation using unique features of isolated human lung radiographs

Abstract

Ex vivo lung perfusion (EVLP) enables advanced assessment of human lungs for transplant suitability. We developed a convolutional neural network (CNN)-based approach to analyze the largest cohort of isolated lung radiographs to date. CNNs were trained to process 1300 longitudinal radiographs from n = 650 clinical EVLP cases. Latent features were transformed into principal components (PC) and correlated with known radiographic findings. PCs were combined with physiological data to classify clinical outcomes: (1) recipient time to extubation of <72 h, (2) ≥ 72 h, and (3) lungs unsuitable for transplantation. The top PC was significantly correlated with infiltration (Spearman R: 0·72, p < 0·0001), and adding radiographic PCs significantly improved the discrimination for clinical outcomes (Accuracy: 73 vs 78%, p = 0·014). CNN-derived radiographic lung features therefore add substantial value to the current assessments. This approach can be adopted by EVLP centers worldwide to harness radiographic information without requiring real-time radiological expertise.

Fig. 1. Class activation mapping for isolated lung images.

Important regions for different predictions are visualized using Grad-CAM. Darker blue colors indicate regions significant for a particular class prediction.

Fig. 2. Correlation heat map depicting the associations between PCs from image features and radiographic scores.

Features analyzed by computer vision are consistent with manually labeled radiographic abnormalities. The shaded bar on the right represents the Spearman correlation coefficients from +1.0 to −1.0. The shade and size of each circle in the grid indicate the extent to which each PC correlated with each radiographic abnormality. (PC principal component, *p < 0.05, **p < 0.01, ***p < 0.001).

Fig. 3. CNN-based image processing pipeline.

1 h and 3 h X-ray images are simultaneously processed by convolutional layers, and one single classification is performed for both images of the same lungs. (CNN convolutional neural network).