Study population

This study conducted a retrospective analysis of 148 percutaneous CT-guided lung biopsies performed at our university hospital from January 2019 to December 2023. We defined a minimum distance of > 10 mm from the pleura to the target lesion for three key reasons. First, this distance enables accurate manual density measurements with minimal standard deviation. Second, it helps exclude inflammatory or infiltrative changes in the lung parenchyma caused by adjacent malignant lesions (safety margin, [40, 41]). Third, this margin facilitates the detection of potential peripheral air trapping [42]. Consecutive exclusion criteria were established to mitigate potential bias from predominant abnormal physiology in the pleural cavity or lung parenchyma (infiltration or effusion). Additionally, factors that could lead to false-positive results in the density measurement, such as interstitial pneumopathy or generalized alveolar patterns in the lung, were excluded. Finally, 72 cases met the inclusion criteria (Fig. 2).

Fig. 2

Flowchart shows the study population

Baseline evaluation

All patients underwent a baseline clinical evaluation, including a medical history review and standard blood tests. Procedure requirements included an international normalized ratio value below 1.5 or a Quick value above 60%, a hemoglobin value exceeding 80 g/L, and a platelet count of over 50 × 109/L, with blood values not older than 5 days. Non-steroidal anti-inflammatory drugs and clopidogrel were discontinued 5 days before the procedure, heparin 6 h before, rivaroxaban 1 day before, and dabigatran and endoxaban 3 days before, following established guidelines.

Biopsy technique

If the biopsy posed a significant risk under normal breathing conditions, it was rescheduled to be performed under anesthesia. Four interventional radiologists, each with 7 to over 10 years of experience, conducted all lung biopsies. The procedures were CT-guided using a Toshiba Asteion 4SL scanner (and either a 17- or 19-gauge coaxial needle paired with an 18- or 20-gauge semiautomated biopsy system (0.828 pitch, 120-kVp tube energy, modulated tube current, and 0.5-s gantry rotation) SemiCut side-cutting only for 18-gauge; Medical Devices Lease S.A., Zug, Switzerland or CorVocetTM full-core; Merit Medical Systems, Utah, United States). Biopsy planning relied on a non-contrast chest CT with 1 mm reconstruction increments (512 × 512-pixel matrix, reconstruction kernel FC7), adhering to the gold standard for needle path planning. Special care was taken to avoid pulmonary vessels, and crossing pulmonary fissures and pleural effusion was forbidden. The interventionalist chose the patient’s positioning based on their experience and the patient’s capabilities. Local anesthesia (1% lidocaine, max 20 mL) was administered. Breathing instructions were omitted to prevent hyperventilation, which could have prolonged the procedure. After tissue sampling, the needle was promptly withdrawn without a sealing agent. A follow-up CT scan was performed immediately post-needle withdrawal, and if no complications arose, the patient was transferred to a supine position on a bed without relocating. If local hemorrhage occurred, another follow-up CT scan was conducted after 5 min. Coughing and hemoptysis were tolerated with stable vital signs, and patients applied counterpressure using the Valsalva maneuver while elevated. A drainage system (Safe-T-Centesis TM 6 or 8F) was utilized if progressive pneumothorax was detected. Patients were monitored for routine vital signs for 4 h post-procedure, and if asymptomatic with no complications, were discharged home. If the control CT scan revealed a non-progressive pneumothorax, a chest X-ray was performed 6 h later for further evaluation. If the pneumothorax exceeded 2 cm, the patient was admitted for one night.

Data collection and image analysis

All procedures were reviewed by a board-certified interventional radiologist with eight years of experience and a radiology resident with three years of experience, both were blinded to the patient’s medical history and did not perform any interventions. All interventional images were analyzed with the software Sectra Workstation (Model IDS7, version 24.2, Patch 4/2022; Sectra AB, Linköping, Sweden).

Radio density measurements

The planning scan reconstructed in lung window (LW) (1 mm reconstruction increments) with a width of 1500 and level of −500 served as the basis. Regions of interest (ROIs) with a diameter of approximately 5 mm are set within the lowest value alongside the access route serving as a reference. Five millimeters was chosen as experience has shown that the lateral error of coaxial needles is usually within this range. The values were recorded in the reconstructed LW and in the MinIP (slap thickness 10 mm) images. Following the findings of Zhang et al [37] we graded our results based on a radiodensity threshold of −850 Hounsfield Units (HU).

To reduce the variability of measured values and compensate for differences in X-ray beam energies [43] and location-related radiodensity variations [15, 44], we applied our Bridged Radiological Observations with a measurement-optimized model (BROM-OLB) for the relative quantitative measurement of radiodensity in CT-guided lung biopsies. The design of this model aimed to generate a reliable depiction of the heterogeneous lung parenchyma, accessible to any interventionalist before biopsy without requiring supplementary software. Three additional ROIs of identical diameter were positioned for comparison with the initial reference ROI, which was placed at the lowest density value along the access route. Their selection was guided by specific criteria to ensure consistency and address potential density variations. Given the height-dependent nature of lung density [45], the first additional ROI was placed at the same vertical level as the ROI along the access route. This positioning aimed to account for variations in density due to gravity and to improve the consistency and reliability of density measurements in different lung regions. The second additional ROI was positioned at an equivalent distance from the pleura, ideally at the same vertical level or within the same third of the lung as the target lesion, to account for density variations from the central to peripheral regions of the lung [46]. The third and final ROI was positioned within visually normal lung parenchyma, either within the same lobe or in a different lung lobe, to serve as a control reference for typical parenchymal density. If the visually normal lung parenchyma had already been measured by the first ROI at the same vertical level, the third ROI was positioned in the other lung. Because the vertical gradient has a more pronounced effect on lung density than the horizontal gradient, it was essential to follow the specified order for setting additional ROIs within this model. Measurements within visually distorted lung parenchyma or ground glass opacity formations were prohibited. This restriction helped to reduce the risk of interference from pathological alterations that could bias density measurements. For our ROI measurements, we selected 1 mm reconstructed images instead of the 5-mm slices commonly used in other studies. This higher-resolution approach provided enhanced spatial detail, allowing for more precise differentiation of small density variations. By reducing the partial volume effect, 1-mm slices offer improved accuracy in capturing subtle density changes across ROI, thereby ensuring the precision necessary for detailed analysis. Additionally, the finer resolution of 1-mm slices enables the accurate positioning of ROIs within clear, homogeneous lung regions, minimizing the risk of artifacts or mixed tissue influences that could otherwise distort density values. To further enhance reliability, we restricted the allowable deviation from the mean measurement value between ROIs to within 20 HU, helping to control for potential attenuation errors while using high-resolution images. A positive rating was assigned if the density of the ROI within the access route was lower than that of most of the other ROIs (the same criterion applied to the minimum intensity projection (MinIP) across the entire slab thickness). In the second step, a MinIP with a slab thickness of 10 mm was generated. Edge sharpening (35%) and contrast enhancement (20%) were added to the MinIP. The ROIs were then re-evaluated and, if necessary, repositioned in more normal lung parenchyma (Fig. 3).

Fig. 3

Technical realization: A native scan is shown in the LW: patient in lateral position and target lesion in the posterior upper lobe segment on the right (arrow). ROI with approx. Five millimeters, and the lowest value in the access route is taken as a reference (R). Three other ROIs of the same size are used for comparison: at the same height (1.H), at the same distance from the pleura (2.PD), and in the parenchyma (3.P) (each in the same or different lung lobe). It is essential to follow the specified order for setting additional ROIs within this model. Measurements within visually destructed lung parenchyma or ground glass opacity formations are prohibited. The deviation from the mean measurement value between the respective ROIs must not exceed 20 HU. Positive rating if the density of the ROI in the access route was lower than the majority of the other ROIs (the same rule also applies in the MinIP over the entire slap thickness). B The same ROIs in the MinIP (slap thickness 10 mm). C After MinIP CE adjustment of the marked ROI (red arrows—shift posteriorly) in visually inconspicuous parenchyma. 2. PD maintains the distance to the pleura from 11 mm. D Access the route with a coaxial needle in front of the target lesion. E Pneumothorax with a seam width of 27 mm and grade 3 lobar hemorrhage. Mittelwert, mean; Abweichung, deviation; Durchmesser, diameter; mm, millimeter

Emphysema, pneumothorax, pulmonary hemorrhage grading, and other investigated parameters

The approved visual assessment system for grading emphysema, according to the Fleischner Society, was used and consequently only recorded in the reconstructed LW images [30]. Binary evaluation of the presence of a pneumothorax on the immediate post-biopsy CT control images, regardless of severity. Pulmonary hemorrhage was evaluated based on the appearance of new consolidative or ground-glass opacity on post-biopsy images and classified using a consensus-based grading system adapted from previous research [16, 18, 47]: Grade 0 indicated no pulmonary hemorrhage, Grade 1 represented needle tract hemorrhage ≤ 2 cm (Fig. 4A, B), Grade 2 indicated hemorrhage > 2 cm but confined to sub lobar regions (Fig. 4C), Grade 3 denoted lobar hemorrhage or larger (Fig. 4D), and Grade 4 indicated hemothorax (Fig. 4E). Other variables assessed included intervention date, birth date, age, sex, procedure time (in minutes), patient position, lesion size (in mm), lesion location, biopsy angle, distance from skin to the lesion (in mm), distance from pleura to lesion (in mm), and lesion location. Needle size, needle system, and the number of samples were recorded from the intervention report. Histological results from the target lesion and the patient’s post-interventional history were collected retrospectively from electronic medical records.

Fig. 4

Pulmonary hemorrhage grading system. CT scan performed before/during (left) and directly after (right) demonstrate the following grades A Grade 0 without image morphological evidence of parenchymal hemorrhage after biopsy. B Grade 1: new linear, fine ground glass with a width under 2 cm can be delineated in the access route after the biopsy. C Grade 2: after the biopsy, the target lesion can no longer be demarcated if the alveolar hemorrhage is locally larger than 2 cm. D Grade 3: lobar hemorrhage and the biopsy channel before. E Grade 4: hemaothorax with immediate hyperdense fluid within the pleural cavity after biopsy. Mittelwert, mean; Durchmesser, diameter; mm, millimeter

Statistical analysis

Statistical analyses were performed using commercially available software (IBM SPSS Statistics for Windows, version 28; IBM, Armonk, NY). A Chi-Square and Fisher exact tests for categorical variables, and the Mann–Whitney U-test for continuous variables. The Kolmogorov-Smirnov test assessed a normal distribution. The Spearman correlation for continuous and contingency correlation for categorical values was used to identify high correlations between the variables. The phi coefficient was calculated for categorical variables and the Pearson correlation coefficient for continuous variables to determine effect sizes. In cases of highly correlated variables, only the variable with the highest effect size was included in the logistic regression model. The optimal measuring method should demonstrate high sensitivity and area under the curve (AUC) while maintaining a low number of false negatives. To evaluate the optimal threshold from the ROC analysis, the Youden Index and the point closest to the top left were used. All tests were performed two-sided with a level of significance p < 0.05. As the side-cutting biopsy system is only available for 18-gauge size (SemiCut, Medical Devices Lease S.A., Zug, Switzerland), homogenization had to be performed for direct comparison of the groups (needle size and system), and cases had to be excluded. To avoid overfitting the regression model, we adhered to the rule of ten [48]. As a result, we included three independent variables in the logistic regression model based on the number of events in the smallest outcome category. For this purpose, we selected the variables that were significant or close to significance, considering a minimum group size of n ≥ 25 for categorical predictors. A logistic regression model was used to assess potential risk factors for pneumothorax and pulmonary hemorrhage [49].