# Evaluation of the impact of cardiopulmonary rehabilitation exercise training on cardiopulmonary function in patients with chronic obstructive pulmonary disease complicated by unstable angina pectoris using a hierarchical deep learning CT image model

## TL;DR

This study shows that cardiopulmonary rehabilitation improves both lung and heart function in patients with COPD and unstable angina, using deep learning to analyze CT scans.

## Contribution

A hierarchical deep learning model is introduced to objectively evaluate CPR effects on cardiopulmonary structure and function in comorbid COPD and UA patients.

## Key findings

- CPR training significantly improved imaging biomarkers of lung, airway, vascular, and heart structures compared to conventional care.
- Clinical parameters like FEV1, FVC, and 6MWD also improved significantly in the CPR group.
- The DL model showed high accuracy in quantifying imaging biomarkers, supporting its use for evaluating rehabilitation outcomes.

## Abstract

This study aimed to quantitatively evaluate the impact of cardiopulmonary rehabilitation (CPR) exercise training on cardiopulmonary structure and function in patients with chronic obstructive pulmonary disease (COPD) complicated by unstable angina (UA) pectoris, based on a hierarchical deep learning (DL) CT image model.

This prospective randomized controlled trial enrolled 400 patients with COPD complicated by UA pectoris, stratified according to GOLD grades (I-IV), who were randomly allocated to an experimental group (EG, n = 200, receiving 12 weeks of standard CPR training) and a control group (CG, n = 200, receiving conventional care). A multi-task 3D U-Net + ResNet50 DL model was constructed to automatically quantify four categories of imaging biomarkers from chest high-resolution CT (HRCT) and coronary CT angiography images: lung parenchyma (percentage of low attenuation volume, LAV%, 15th percentile density Perc15, mean lung density, MLD), airways (percentage of airway wall thickness WA%, Pi10), pulmonary vasculature (percentage of blood vessels <5 mm in cross-sectional area BV%, vascular fractal dimension), and heart (coronary artery calcium score, CACS, left ventricular mass, LVM, ejection fraction, EF, stroke volume, SV). Concurrently, pulmonary function, cardiopulmonary exercise testing parameters, and 6-min walk distance (6MWD) were assessed at baseline, 6 weeks, and 12 weeks of the intervention.

Versus CG, after 12 weeks of intervention, EG demonstrated notable imaging improvements across all GOLD grades: decreased LAV%, increased Perc15 and MLD; reduced WA% and Pi10; increased BV5% and vascular fractal dimension; improved EF and SV, and decreased LVM (all P < 0.05). Clinically, EG also showed substantially better FEV1, FVC, peak VO2, and 6MWD than CG (P < 0.01). Correlation analysis revealed moderate to strong correlations between these imaging metrics and clinical functional parameters (|r| = 0.36–0.62, P < 0.001). The constructed DL model demonstrated excellent segmentation accuracy (Dice coefficient: 0.87–0.95) and quantification reliability in both internal and external validations.

CPR not only substantially improves clinical cardiopulmonary function but also induces beneficial structural remodeling of the lung parenchyma, airways, pulmonary vasculature, and heart in patients with COPD complicated by UA pectoris. The DL-based CT image quantification framework provides reliable imaging biomarkers for objectively evaluating rehabilitation efficacy in these comorbid patients, offering new evidence for personalized management of cardiopulmonary comorbidities.