Predicting Postoperative Lung Function in Patients with Lung Cancer Using Imaging Biomarkers

There have been previous studies conducted to predict postoperative lung function with pulmonary function tests (PFTs). Computing tomography (CT) can quantitatively measure small airway walls’ thickness, lung volume, pulmonary vessel volume, and emphysema area, which reflect the severity of respiratory diseases. These measurements are considered imaging biomarkers. This study aimed to predict postoperative lung function with imaging biomarkers. A retrospective analysis of 79 patients with lung cancer who had undergone lung surgery was completed. Postoperative lung function measured by forced expiratory volume in one second (FEV1) was defined as an outcome. Preoperative clinico-pathological parameters and imaging biomarkers representing airway walls’ thickness, severity of emphysema, total lung volume, and pulmonary vessel volume were measured quantitatively in chest CT by an automated segmentation software, AVIEW COPD. Pi1 was defined as the first percentile along the histogram of lung attenuation that represents the degree of emphysema. Wafw was defined as the airway thickness, which was calculated by the full-width at half-maximum method. Logistic and linear regressions were used to assess these variables. If the actual postoperative FEV1 was higher than the postoperative FEV1 projected by a formula, the group was considered to be preserved. Among the 79 patients, 16 of the patients were grouped as a non-preserved group, and 63 of them were grouped as a preserved group. The patients in the preserved FEV1 group had a higher vessel volume than the non-preserved group. Pi1 and Wafw were independent predictors of postoperative lung function. Imaging biomarkers can be considered significant variables in predicting postoperative lung function in patients with lung cancer.


Introduction
Chronic obstructive pulmonary disease (COPD) is a major independent risk factor for lung cancer and a common comorbid disease [1].COPD is characterized by persistent airflow limitation, and a pulmonary function test (PFT) is required to make the diagnosis.The severity of airflow limitation in patients with COPD is evaluated by post-bronchodilator forced expiratory volume in one second (FEV 1 ) [2].However, patients with the same stage of severity based on FEV 1 have different degrees of emphysema.Due to this limitation and the development of computing tomography (CT) analysis techniques, although PFT has been previously used as a method to evaluate lung function, research using CT as a method to understand the pathophysiology and quantitatively measure the severity of COPD to treat and predict its exacerbations has increased over the past decade [3].
Surgery is the standard therapy for early-stage lung cancer, and postoperative lung function is crucial because it is related to mortality.Therefore, predicting postoperative lung function is important in order to decide treatment modalities.Postoperative FEV 1 is a variable widely used to represent postoperative lung function.Since patients with lung cancer have COPD as a common comorbid disease, they usually have a limitation of the airways, resulting in a decrease in FEV 1 .Previous studies have estimated predicted postoperative FEV 1 (ppoFEV 1 ) using the following formula: ppoFEV 1 = preoperative FEV 1 × (19-segments to be removed)/19 [4,5].A meta-analysis has shown that CT volume and density measures are the preferred method for predicting postoperative FEV 1 [6].In our previous study [7], we showed that preoperative residual volume (RV)/total lung capacity (TLC) is positively correlated with postoperative lung function.As a result, in addition to the number of segments removed, lung function should also be taken into account because the concurrent respiratory condition affects postoperative lung function.Conventionally, PFT has been used as a method to evaluate lung function, but there is growing evidence that imaging modalities such as CT can also estimate lung function [6].
Due to the development of deep learning, the airway tree segmentation method in chest CT has been improved [8].After airway segmentation, it is feasible to evaluate the emphysema region by calculating the low-attenuation area, measuring the thickness of the airway wall, the total lung volume, and the total vascular volume.These variables are considered imaging biomarkers.The aim of this study was to determine whether imaging biomarkers can be used as variables in predicting postoperative lung function in patients with lung cancer.

Patients
We retrospectively reviewed 94 patients with lung cancer who had undergone lung surgery at Eunpyoung St. Mary's Hospital between July 2016 and February 2021.Four patients were excluded due to missing PFT values.Eleven patients were excluded because their CT scans were not compatible with the software used in this study.Thus, a total of 79 patients were finally included in this study.

AI Algorithm
A deep learning algorithm (AVIEW COPD, Coreline Soft, Seoul, South Korea), available commercially, was utilized to examine the imaging biomarkers obtained from the CT scans.The algorithm was implemented to perform airway segmentation using the 2.5D convolutional neural net (2.5D CNN), which was trained and evaluated by CT scans obtained from the Korean obstructive lung disease study [8].This automated segmentation software significantly improved branch level segmentation accuracy [9].

Chest CT Acquisition
The CT scanners used in this study were CT scanners manufactured by Seimens Healthineers (Somatom Definition Edge; Forchheim, Germany), GE Healthcare (Revolution ACT; Chicago, IL, USA), and Philips (Brilliance 64 CT; Amsterdam, The Netherlands).The parameters of the Somatom Definition Edge scanner were 90-130 kVP, 65-308 mAs, and a slice thickness of 3 mm.The parameters of the Revolution ACT scanner were 100 and 120 kVP, 1-12 mAs, and a slice thickness of 1.25 and 2.5 mm.The parameters of the Brillance 64 CT machine were 120 kVP, 166-172 mAs, and a slice thickness of 5 mm.

Variables' Explanation and Outcome Definition
The patients' age, sex, pathology type, tumor stage according to the eighth tumor node metastasis (TNM) classification and location, operation type, treatment modalities, and preoperative PFT values, including FEV 1 , FEV 1 /FVC, RV/TLC, and diffusing capacity of the lung for carbon monoxide (DL CO ), were collected.The patients were classified as non-smokers if they had never smoked or had smoked less than 100 cigarettes in their lives and as ever smokers if they had smoked more than 100 cigarettes.All the PFT values were measured before the inhalation of a bronchodilator.The operation type was divided into two groups-video-assisted thoracic surgery (VATS) and open surgery.Neoadjuvant chemotherapy, adjuvant chemotherapy, palliative chemotherapy, neoadjuvant radiotherapy, adjuvant radiotherapy, and palliative radiation were all used as treatment methods.In this study, we considered imaging biomarkers, which are quantitatively measured in chest CT, as the variables.Emphysema, airway thickness, total lung volume, and pulmonary vessel volume were measured quantitatively in the chest CT using an automatic segmentation software (AVIEW COPD, Corelinesoft, Seoul, Republic of Korea).
Pulmonary emphysema is defined as an abnormal permanent enlargement of airspaces distal to the terminal bronchioles [10,11] and results in low-attenuated areas (LAA) in a chest CT that are below −950 Hounsefield unit (HU) [3].Since the extent of emphysema in CT is associated with pulmonary function decline [12], the variables representing emphysema were evaluated.The percentage of LAA compared to the whole lung measured in cubic centimeters (cc) was defined as the variable LAAsize for the quantification of emphysema.Two variables, Pi1 and Pi15, which correspond to the first and fifteenth percentiles along the histogram of lung attenuation, were defined as an alternative method to quantify emphysema [13].
The remodeling of small airways, which have an internal diameter smaller than 2 mm, is known to represent airflow limitation in patients with COPD.Through the remodeling process of these airways, the airways become thicker, and the thickness of the airways is strongly associated with the progression of COPD [14].However, due to the limited resolution of CT scans, small airways are difficult to measure.Nakano et al. overcame this obstacle by showing that relatively larger airways measured by CT reflect small airways measured histologically [15].In order to represent airway thickness, three variables-Wafw, Waband, and Pi10fw-were measured.The full-width half-maximum (FWHM) method is the most frequently used technique to measure airways quantitatively.The inner and outer margins of the airway wall were measured using linear rays from the airway center.The airway thickness was measured by the difference between the half point between the minimum and maximum gray levels and the maximum point of the gray level [16,17].The proportion of the mean of the airway wall area calculated by the FWHM method was defined as Wafw.However, the FWHM method overestimates the wall dimensions in small airways [17].The integral-based half-band (IBHB) method, which is a threshold-based method, was developed and showed better correlation with pulmonary functions [18].The mean of the airway wall area calculated by the IBHB method was defined as Waband.Another parameter, Pi10, was developed.It was derived by plotting the square root of the airway wall area against the internal perimeter of each measured airway.After creating a regression line, Pi10 was defined as the square root of the airway wall area with an internal perimeter of 10 mm on the regression line [19].The airway wall area was calculated by the FWHM method, and the corresponding Pi10 value was defined as Pi10fw.
The total lung volume, measured in cc, was defined as TLV.The total airway count, which quantifies the total number of visible airways from the pulmonary tree, was defined as TAC.It is well known that pulmonary lung function and pulmonary vasculature are correlated.Park et al. showed that, as emphysema severity increased, the number of pulmonary vessels decreased [20].The total vessel volume, which quantifies the total vessel volume, including pulmonary arteries and pulmonary veins, was defined as VV.
The pre-bronchodilator value of FEV 1 (%) measured at 6 ± 3 months after surgery was defined as the outcome.In order to analyze the characteristics of patients who have a better lung function than predicted after surgery, the patients were categorized into two groups.The patients were grouped into the FEV 1 -preserved group if they had a higher FEV 1 than the ppoFEV 1 and into the FEV 1 -non-preserved group otherwise.Linear and logistic regressions were performed to analyze the variables.p values < 0.05 were indicated as showing statistical significance.Statistical analyses were performed using R (version 4.0.2;The R foundation, Vienna, Austria).

Overall Patient Characteristics
A total of 79 patients with primary lung cancer who had undergone surgery with mediastinal lymph node dissection were included in this study.The patients' characteristics are presented in Table 1.The mean age of the patients was 69.18 years, and the majority of the patients were male (64.56%).A total of 36 patients (45.57%) were never smokers, and 43 patients (54.43%) were smokers.Regarding the operation method, 26 patients (32.91%) received open surgery, and 53 patients (67.09%) received VATS.Stage I was the most common stage (78.48%), and the right upper lung was the most common site of the cancer (34.18%).About 22.78% of the patients had received adjuvant chemotherapy.The preoperative PFT results were measured before bronchodilator usage, and the imaging biomarkers were measured using the automatic software AVIEW COPD.capacity; Pi10fw, internal perimeter of the segment of the bronchus measured in the full-width at half-maximum method; LAAsize, low-attenuation area size; Pi1, first percentile Hounsefield unit of the low-attenuation area of the lung; Pi15, fifteenth percentile Hounsefield unit of the low-attenuation area of the lung; TAC, total airway count; VV, total vessel volume; and TLV, total lung volume.

Comparison between FEV 1 -Non-Preserved Groups and -Preserved Groups
The comparison between the FEV 1 -non-preserved and -preserved groups are provided in Table 2.The age, sex, histologic features, smoking history, operation type, stage, location, and preoperative PFT values did not differ between the two groups.The mean value of the total vessel volume was significantly higher in the FEV 1 -preserved group than in the non-preserved group (113.76 ± 46.7 and 89.99 ± 33.45, respectively with p value 0.027).

Factors Associated with Preserved Postoperative FEV 1
Table 3 displays the findings of the logistic regression analysis that determined the variables linked to postoperative FEV 1 preservation.The outcome was defined as positive if the postoperative FEV 1 was higher than the ppoFEV 1 , which means that it is "preserved" and is negative otherwise.Adjuvant chemotherapy (unadjusted odds ratio (OR) 0.392, 95% confidence interval (CI) 0.106-9.779,p = 0.124), Pi10fw (unadjusted OR 0.379, 95% CI 0.086-1.673,p = 0.2), TAC (unadjusted OR 1.006, 95% CI 0.997-1.016,p = 0.176), and VV (unadjusted OR 1.015, 95% CI 0.999-1.031,p = 0.063) had p values less than 0.2 in the univariate analysis that were entered into the multivariate analysis.None of the variables were statistically significant in the multivariate logistic regression analysis.
The results of the linear regression are provided in Table 4. Since age is a strong factor which is associated with postoperative lung function, it was entered into multiple linear regressions with variables that had p values less than 0.2.The multiple linear regression model had an R 2 value of 0.134 and a p value of 0.024.Pi1 and Wafw remained statistically significant.Pi1 was positively correlated, and Wafw was negatively associated.

Comparison of Conventional Formula and Multiple Linear Regression Model
Residual box plots of the conventional formula model and the multiple linear regression model are presented in Figure 1.The residual was defined as the difference between the predicted and the actual postoperative FEV 1 .In the conventional formula model, the ppoFEV 1 was calculated using the conventional formula.In the multiple linear regression model, the ppoFEV 1 was calculated using the following formula: ppoFEV 1 = 373.8+ 0.28 × age + 0.156 × Pi1 − 2.111 × Wafw − 0.001 × TLV.The value of intercept in this model was 373.8.

Discussion
In this study, we used imaging biomarkers to predict postoperative lung function.The total vessel volume, including pulmonary arteries and vessels, was higher in the FEV1preserved group.The Pi1, the first percentile of the low-attenuating area, was positively correlated, and the wall area (% ratio of the wall area to the airway area) measured by the FWHM method was negatively correlated with the postoperative FEV1.We developed a novel formula for multivariate linear regression utilizing these variables.We compared this formula with the conventional formula, which uses the number of lung segments removed by the residual box plot.The residual was defined as the difference between the actual postoperative FEV1 and the formula-predicted postoperative FEV1.The residual mean and the interquartile box were closer to zero in our new formula, showing the promising effects of radiological biomarkers in predicting postoperative lung function.
Approximately 40-70% of patients with lung cancer have coexisting COPD [1].An important pathophysiological characteristic of COPD is pulmonary vascular alteration.In the patients with severe COPD, there was a decreased ratio of the cross-sectional area of tiny pulmonary arteries to the overall area of the lung [21].As the severity of emphysema based on chest CT imaging increased, the number of pulmonary vessels decreased [20].A previous study has shown the pathophysiology of this pulmonary vascular change.In said study, the vascular endothelial growth factor (VEGF), which induces endothelial cell growth, and the vascular endothelial growth factor receptor 2 (VEGFR2) were significantly decreased in emphysematous lungs [18].Our study was consistent with these previous studies.The patients in the FEV1-non-preserved group had lower total pulmonary vessel volumes, measured with the AVIEW COPD system.This indicated that patients with a severe form of emphysema tend to have a lower postoperative lung function than

Discussion
In this study, we used imaging biomarkers to predict postoperative lung function.The total vessel volume, including pulmonary arteries and vessels, was higher in the FEV 1preserved group.The Pi1, the first percentile of the low-attenuating area, was positively correlated, and the wall area (% ratio of the wall area to the airway area) measured by the FWHM method was negatively correlated with the postoperative FEV 1 .We developed a novel formula for multivariate linear regression utilizing these variables.We compared this formula with the conventional formula, which uses the number of lung segments removed by the residual box plot.The residual was defined as the difference between the actual postoperative FEV 1 and the formula-predicted postoperative FEV 1 .The residual mean and the interquartile box were closer to zero in our new formula, showing the promising effects of radiological biomarkers in predicting postoperative lung function.
Approximately 40-70% of patients with lung cancer have coexisting COPD [1].An important pathophysiological characteristic of COPD is pulmonary vascular alteration.In the patients with severe COPD, there was a decreased ratio of the cross-sectional area of tiny pulmonary arteries to the overall area of the lung [21].As the severity of emphysema based on chest CT imaging increased, the number of pulmonary vessels decreased [20].A previous study has shown the pathophysiology of this pulmonary vascular change.In said study, the vascular endothelial growth factor (VEGF), which induces endothelial cell growth, and the vascular endothelial growth factor receptor 2 (VEGFR2) were significantly decreased in emphysematous lungs [18].Our study was consistent with these previous studies.The patients in the FEV 1 -non-preserved group had lower total pulmonary vessel volumes, measured with the AVIEW COPD system.This indicated that patients with a severe form of emphysema tend to have a lower postoperative lung function than expected.
Emphysema replaces the normal lung with air-containing spaces, resulting in CT attenuation.A value of -950 HU is usually used as the threshold to identify emphysema lesions.To avoid noise in CT measurements and obtain robust results, the percentile approach was introduced as an alternative.Emphysema quantification was measured by CT attenuation at a given percentile along the histogram of lung attenuation [3,22].In our study, Pi1, the first percentile of the low-attenuating area, was positively correlated with postoperative lung function.Since the severity of emphysema was lower in the patients with a higher Pi1, the patients with less-severe emphysema had a higher postoperative lung function.
The progression of COPD is associated with inflammatory mucous exudates in the lumen and infiltration of the wall by inflammatory immune cells resulting in the thickening of the airway's walls [14].Han et al. showed that a greater airway wall thickness is associated with COPD exacerbation frequency [23].Wafw, an imaging biomarker representing airway wall thickness, was negatively correlated with postoperative lung function in our study.This result was consistent with previous studies [3].Waband and Pi10fw, two further radiological indicators of airway wall thickness, failed to reach statistical significance.
The total lung volume in patients with COPD is increased due to expiratory airflow limitation [24].The total lung volume was found to be inversely linked with the outcome in a single linear regression analysis, meaning that individuals with greater lung volumes had a lower postoperative lung function than expected (p = 0.159).However, the total lung volume was not statistically significant in multiple linear regressions (p = 0.656).
Overall, the multiple linear regression model with imaging biomarkers showed a better performance than the conventional formula in terms of predicting postoperative lung function.Widely used conventional formulas use preoperative FEV 1 and the number of segments resected as the variables [25].PFT results, including FEV 1 , are routinely used to evaluate lung function, but there are some limitations.Elderly patients or patients with a general weakness might not perform PFT correctly [26].The volume of each lung segment can also differ.Contrary to this variability, imaging biomarkers are quantified in chest CT, which offers repeatable and reliable data, with reduced variations, for those who are unable to perform PFT.
Regarding imaging biomarkers, airway tree segmentation is important because it quantifies anatomical features.In our study, segmentation was performed via a semiautomated method, using AVIEW COPD.Segmentation was performed voxel-by-voxel by a 2.5D CNN and trained in a supervised manner.The 2.5D patches captured the 3D structure of the airway tree efficiently.Since this method has been validated on multiple datasets, these imaging biomarkers are highly reliable [8].Therefore, imaging biomarkers can be considered important factors in predicting postoperative lung function.
However, our study has several limitations.First, the number of patients enrolled was relatively small.Second, excluding patients without postoperative FEV 1 might have led to selection bias.Patients with postoperative FEV 1 may present substantially reduced lung function since PFT is not frequently performed, and physicians typically only perform PFT when patients have symptoms.Future studies are required with a large number of patients, and PFT should be performed routinely to avoid selection bias.

Conclusions
Imaging biomarkers can be considered significant variables in predicting postoperative lung function in patients with lung cancer.The total pulmonary vessel volume was higher in the lung function-preserved group.Pi1, an imaging biomarker representing emphysema, was positively correlated, and Wafw, an imaging biomarker representing airway wall thickness, was negatively correlated with postoperative lung function.

Diseases 2024 , 11 Figure 1 .
Figure 1.Comparison of conventional formula and multiple linear regression model in predicting postoperative lung function, shown with residual box plots.

Figure 1 .
Figure 1.Comparison of conventional formula and multiple linear regression model in predicting postoperative lung function, shown with residual box plots.

Table 1 .
Clinical characteristics of the study patients.

Table 3 .
Factors associated with preserved postoperative FEV1 using logistic regression.
OR; odds ratio; CI; confidence interval; Sqcc, squamous cell carcinoma; BUL, both upper lobes; V0, preoperative; FEV1, forced expiratory volume in one second; FVC, forced vital capacity; DLco, diffusing capacity of the lung for carbon monoxide; RV, residual volume; TLC, total lung capacity; Pi10fw, internal perimeter of the segment of the bronchus measured in the full-width at half-maximum method; LAAsize, low-attenuation area size; Pi1, first percentile Hounsefield unit of the low-attenuation area of the lung; Pi15, fifteenth percentile Hounsefield unit of the low-attenuation area of the lung; TAC, total airway count; VV, total vessel volume; and TLV, total lung volume.

Table 4 .
Factors associated with postoperative FEV 1 using linear regression.
one second; FVC, forced vital capacity; DLco, diffusing capacity of the lung for carbon monoxide; RV, residual volume; TLC, total lung capacity; Pi10fw, internal perimeter of the segment of the bronchus measured in the full-width at half-maximum method; LAAsize, low-attenuation area size; Pi1, first percentile of the Hounsefield unit of the lung; and Pi15, fifteenth percentile of the Hounsefield unit of the lung.