Direct assessment of lung function in COPD using CT densitometric measures

Abstract

Purpose To investigate whether lung function in patients with chronic obstructive pulmonary disease (COPD) can be directly predicted using CT densitometric measures and assess the underlying prediction errors as compared with the traditional spirometry-based measures. Materials and Methods A total of 600 CT examinations were collected from a COPD study. In addition to the entire lung volume, the extent of emphysema depicted in each CT examination was quantified using density mask analysis (densitometry). The partial least square regression (PLS regression) was used for constructing the prediction model, where a repeated random split-sample validation was employed. For each split, we randomly selected 400 CT exams for training (regression) purpose and the remaining 200 exams for assessing performance in prediction of lung function (e.g., FEV1 and FEV1/FVC) and disease severity. The absolute and percentage errors as well as their standard deviations were computed. Results The averaged percentage errors in prediction of FEV1, FEV1/FVC%, TLC, RV/TLC% and DLco% predicted were 33%, 17%, 9%, 18% and 23%, respectively. When classifying the exams in terms of disease severity grades using the CT measures, 37% of the subjects were correctly classified with no error and 83% of the exams were either correctly classified or classified into immediate neighboring categories. The linear weighted kappa and quadratic weighted kappa were 0.54 (moderate agreement) and 0.72 (substantial agreement), respectively. Conclusion Despite the existence of certain prediction errors in quantitative assessment of lung function, the CT densitometric measures could be used to relatively reliably classify disease severity grade of COPD patients in terms of GOLD.

Predicting Pulmonary Function Testing from Quantified Computed Tomography Using Machine Learning Algorithms in Patients with COPD

“…Nevertheless, Gu et al predicted lung function values based on the emphysema acquired by qCT. They calculated MREs in the prediction of the FEV 1 % between 17% and 256% [32]. Similar to our study, the prediction of FEV 1 /VC performed better when compared with the FEV 1 % prediction.…”

“…Further evaluations with larger sample sizes might benefit the XGBoost and neural-network-based prediction. Nevertheless, the prediction performance of the models used in this work significantly overcomes previous, one-linear regression systems using explorations [32].…”

A Feature-Based Approach to Big Data Analysis of Medical Images

“…A mixture density is estimated for each input feature of a new image as the weighted sum of 1) a variable bandwidth kernel density computed from a set of nearest neighbors and 2) a background distribution over unrelated features. This estimator achieves state-of-the-art performance in automatically predicting the 5-class GOLD severity label from a large multi-site data set, improving upon previous results based on lung-specific image processing pipelines and single-site data [13, 28, 29]. Furthermore, subject-level classification is used to identify all instances duplicate subjects in a large set of 19,000 subjects, despite deformation due to breathing state.…”

“…Mets et al [28] use densitometric measures computed from single-site data of 1100 male subjects, to achieve an AUC value of 0.83 for binary COPD classification. Gu et al [29] use automatic lung segmentation and densitometric measures to classify single-site data according to the GOLD range, achieving an exact classification rate of 0.37, or 0.83 if classification into neighboring categories is considered correct. A major challenge is classifying multi-site data acquired across different sites and scanners.…”

A Sensor for Spirometric Feedback in Ventilation Maneuvers during Cardiopulmonary Resuscitation Training

“…Noninvasive methods used to characterize human and animal respiration require advanced techniques and are costly, such as: computed tomography and densitometry [1,2,3,4,5,6], electrical impedance tomography [7,8], magnetic resonance imaging [9,10,11], contrast radiology [12], image ultrasound [13,14], ultrasonic sensors [15,16,17], pulse oscillometry [18,19], electrostatic methods [20], closed circuits with inert gases [21], impedance pneumography and plethysmography [22,23] and electronic noses [24].…”