# Volume histogram analysis of whole-lung CT: differentiating usual from nonspecific interstitial pneumonias and predicting prognosis

## TL;DR

A new CT-based model using volume histogram analysis can distinguish UIP from NSIP and predict patient outcomes in interstitial pneumonia.

## Contribution

A formula-based VHA model was developed to differentiate UIP from NSIP with strong diagnostic performance and prognostic correlation.

## Key findings

- The VHA model achieved an AUC of 0.81 in external validation for differentiating UIP from NSIP.
- VHA model outcomes were significantly correlated with patient prognosis (hazard ratio 1.60).

## Abstract

Low agreements among experts for differentiating usual interstitial pneumonia (UIP) from nonspecific interstitial pneumonia (NSIP) motivate the use of automated imaging diagnosis. Volume histogram analysis (VHA) of the lung parenchyma using computer-aided diagnostic software is more straightforward to perform and interpret than radiomics. To assess whether a predictive model generated by VHA (VHA model), using voxel data of each lung lobe obtained via whole-lung CT, can differentiate radiological UIP from NSIP, and to explore the relationship between VHA model outcomes and patient prognosis.

This study included 74 patients from one university hospital (cohort A: 47 patients with idiopathic pulmonary fibrosis [IPF]/UIP and 27 with idiopathic NSIP [iNSIP] and connective tissue disease-associated NSIP [CTD-NSIP]) and 146 patients from another hospital (cohort B: 111 with IPF/UIP and 35 with iNSIP/CTD-NSIP), with diagnoses confirmed through multidisciplinary discussion. Using the VHA values obtained from each lung lobe in cohort A, a formula-based VHA model was developed. The regularization parameters were optimized using five-fold cross-validation to maximize the area under the receiver operating characteristic curve (AUC). This VHA model was externally validated in cohort B. The correlation between various parameters and prognosis was analyzed using Cox proportional hazards multivariate analysis.

The mean AUC of the best VHA model that differentiated UIP patterns in cohort A was 0.91 (95% confidence interval [CI], 0.84–0.98), with a positive predictive value (PPV) of 0.97 (0.88–1.00). External validation of this model for cohort B revealed that the AUC for UIP differentiation was 0.81 (0.70–0.88), with a PPV of 0.94 (0.88–0.98). Multivariate analysis revealed that the values calculated by the VHA model were correlated with prognosis (hazard ratio, 1.60; 95% CI, 1.17–2.18; p = 0.003).

The VHA model could effectively differentiate radiological UIP patterns and may help predict the prognosis of patients with interstitial pneumonia.

A formula-based model using CT volume histogram analysis (VHA) of each lung lobe was developed to differentiate usual interstitial pneumonia (UIP) from nonspecific interstitial pneumonia (NSIP). The VHA model demonstrated strong diagnostic performance, achieving an area under the curve of 0.81 in external validation, and also statistically correlated with patient prognosis.

The online version contains supplementary material available at 10.1007/s11604-025-01880-9.