RPS 804 - Current topics in lung cancer imaging

Author Block: Y. Xin¹, S. Z. Erick¹, M. A. Heuvelmans¹, G. De Bock¹, R. Vliegenthart¹, B. Edwin², M. Van Tuinen¹, F. Mohamed Hoesein²; ¹Groningen/NL, ²Utrecht/NL
Purpose: CT-based body composition measures improve prediction of mortality in oncology, but reference values are lacking. We evaluated age- and smoking-related variation in body composition in a lung cancer screening cohort.
Methods or Background: We used artificial intelligence-based automated analysis of body composition measures on baseline low-dose chest CT from male participants in the NELSON lung cancer screening trial. Skeletal muscle area (SMA) and subcutaneous adipose tissue area (SAT) were quantified at T5, T8, and T10 levels. The mean values of these levels were used to derive a single SMA and SAT metric per participant; the fat-to-muscle ratio (FMR) was calculated. Age in 5-year groups and smoking status were analyzed in relation to body composition measures. Smoking pack-years was included as covariate in regression analyses.
Results or Findings: We included 4,435 men with mean age 59.4 (SD=5.6) years and mean smoking pack-years 42.2 (SD=29.7). Current smokers (55.0%%) had significantly lower SAT (372 vs. 441 cm², p<0.001), SMA (501 vs. 507 cm², p<0.001) and FMR (0.74 vs. 0.87, p<0.001), compared with former smokers(45.0%). Across 5-year age groups, SMA declined from 515 cm² in the 50–54 years group to 472 cm² in the ≥70 years group (p<0.001), while SAT increased from 376 to 443 cm² (p<0.001) and FMR from 0.70 to 0.90 (p<0.001); associations remained significant after adjusting for smoking status and pack-years.
Conclusion: In men undergoing lung cancer screening, higher age was associated with muscle loss and fat gain, while current smoking was associated with lower muscle and fat compared to former smoking. Reference values of chest CT-based body composition measures may help risk stratification in screening.
Limitations: Our further analyses should confirm the relationship of body composition markers to outcomes in lung cancer screening setting.
Funding for this study: Funding was provided by the Dutch Cancer Society and Siemens Healthineers.
The first author is supported by funding from the Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, China.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Health Council, WBO Committee. Population Screening Act: CT screening on lung cancer. The Hague: Health Council of the Netherlands, 2000.

Author Block: E. Cavigli¹, G. Picozzi¹, D. Puliti¹, F. Cortez Ibanez², G. R. De Luca³, V. Miele¹, S. Delorme², R. Kaaks², M. Mascalchi¹; ¹Florence/IT, ²Heidelberg/DE, ³Bologna/IT
Purpose: To determine the frequency and prognosis of lung cancer associated with cystic airspaces (LCCA) in LDCT screening, and to evaluate the potential impact of Lung-RADS v2022 on earlier detection
Methods or Background: LCCAs were identified by two experienced radiologists who jointly reviewed LDCT examinations from 714 screen-detected and 463 non-screen-detected lung cancers from the NLST, ITALUNG, and LUSI trials. Radiological features were correlated with histology, staging, and survival. Lung-RADS v2022 criteria were retrospectively applied. Statistical analyses included chi-square tests and Cox regression models
Results or Findings: Seventy-three LCCAs were identified: 44 among screen-detected cases (6.1%) and 29 (3 interval and 26 post-trial) among non-screen-detected cases (6.2%). Retrospective review revealed that 20/29 (68.9%) of non-screen-detected LCCAs were visible in prior LDCTs. Morphological patterns included unilocular cysts with mural nodules (35/73, 41%), circumferential solid walls (19/73, 26%), and multilocular cysts (24/73, 33%). Histology revealed adenocarcinoma/BAC in 50/73 cases (68.5%), squamous/adeno-squamous carcinoma in 16 (21.9%), and unclassified carcinoma in 7 (9.6%). Five years after diagnosis, 35 patients (47.9%) had died. Mortality was significantly higher for non-screen-detected (p=0.012) and stage III-IV cases (p<0.001). Lung-RADS v2022 criteria could have led to earlier diagnostic work-up in 62/73 (84.9%) of cases
Conclusion: LCCAs account for approximately 6% of lung cancers in LDCT screening. Screen detection and early-stage diagnosis are associated with better prognosis. Application of Lung-RADS v2022 recommendations could facilitate earlier identification and management of LCCA.
Limitations: Retrospective investigation; we didn't assess whether the lesion sizes were predictive of the survival; we didn't assess the frequency in our screening cohorts of benign cysts not evolving into LCCA; we didn't evaluate neither the reproducibility nor the specificity of the Lung-RADS v2022 criteria
Funding for this study: No
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: G. Chassagnon, A. Lupo, M. ROYER, M. Alifano, N. Roche, M-P. Revel; Paris/FR
Purpose: Preoperative risk stratification before lung cancer surgery mainly relies on clinical indicators. This study aimed to determine whether chest computed tomography (CT)-derived parameters could provide additional prognostic value in identifying patients at increased risk of peri-operative mortality.
Methods or Background: We retrospectively analyzed data from 720 patients who underwent lung cancer resection at our tertiary center in 2018-2019. Clinical, functional and imaging data were analyzed, including for the latter, coronary artery calcium score, pulmonary artery and aorta diameters, presence of emphysema, interstitial lung abnormalities, or airspace enlargement with fibrosis (AEF). Multivariable logistic and Cox regression models were applied to identify independent predictors of 30-day and 90-day mortality, as well as overall survival.
Results or Findings: AEF on CT (Odds Ratio (OR) 10.87, p < 0.001) was a significant predictor of 30-day mortality, while AEF (OR 7.04, p<0.001), as well as a higher ECOG status (OR 1.98, p=0.029) and a lower FEV₁ (OR 0.97, p=0.023) were associated with increased 90-day mortality. AEF was present in 8.2% of patients and was observed in 45.5% of patients who died within 30 days postoperatively. AEF was a nearly significant predictor of long-term survival in the multivariate Cox regression (HR = 1.57, 95% CI: 0.99–2.47, p = 0.046).
Conclusion: CT can provide valuable prognostic information in patients undergoing lung cancer surgery. AEF on CT is a major independent prognostic marker, especially for perioperative mortality. Our results support recognizing AEF as a distinct radiological entity that should be systematically assessed.
Limitations: Firstly, AEF was defined radiologically rather than histologically. Histological confirmation of AEF was unavailable because the affected lung areas were not systematically sampled and analysed.. Secondly, it was a single-centre retrospective study, which may limit the generalizability of the findings.
Funding for this study: This study received no funding



This study received non funding
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study protocol was approved by the local Institutional Review Board (CLEP Decision N°: AAA-2025-10005), which waived the need for patient consent.

Author Block: M. F. G. Konrad¹, E. Nischwitz¹, J. Chorostowska-Wynimko², J. Vogel-Claussen³, J. Moes-Sosnowska², M. Adamek⁴, A. Kerpel-Fronius⁵, H. Prosch⁶, H-U. Kauczor¹; ¹Heidelberg/DE, ²Warsaw/PL, ³Hannover/DE, ⁴Gdańsk/PL, ⁵Budapest/HU, ⁶Vienna/AT
Purpose: Assessing the current status of applied CT image acquisition protocols in lung cancer screening (LCS) worldwide, focusing on technical factors linked to radiation exposure.
Methods or Background: The survey was expanded from an internal SOLACE version to include broader dissemination, collecting data from personnel responsible for the definition of CT protocols at LCS centres worldwide. Data were collected through a baseline survey between June 2024 and September 2025 to represent the most current status.
Results or Findings: Survey responses were received from 71 screening centres from 29 countries (19 Europe, 10 other continents). Institutional factors influencing CT protocols encompass the responsibility of establishment and modification of protocols by personnel (radiologists, radiographers, medical physicists, manufacturer personnel). In 33% of the institutions the protocol was established by a multiprofessional team. CT protocols were mostly modifiable (85%). Technical questions were partially answered. In 88% of the centres (43 of 49) automatic exposure control was implemented. Reconstructed slice thickness ranged 0.625-1.5 mm; 1.0 mm dominating with 64%. Increment ranged 0.5-1.25 mm, where 0.625 mm, 0.7 mm, and 1.0 mm were distributed more evenly with 18%, 25%, and 30%, respectively. Screening-specific software was used at 87% of sites; mainly for nodule detection, volumetry, and calculation of volume doubling time (32, 31, and 25 centres, respectively). Reconstruction algorithm types included filtered-back projection (4), iterative reconstruction with statistical modeling (30), and iterative reconstruction with deep learning support (7).
Conclusion: LCS imaging often reaches the technical limits of currently operated devices. Multiprofessional establishment of CT protocols is an area for improvement. Variations in reconstruction algorithm types warrant further research regarding their influence on volumetry calculations. Protocol optimization is essential to balance radiation exposure reduction and diagnostic quality.
Limitations: The survey likely reflects centres with research interests.
Funding for this study: This project is co-funded under the EU4Health Programme 2021–2027 under grant agreement no. 101101187
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: F. Palmeri¹, A. Del Gaudio², M. Zerunian², F. Di Gregorio², A. Ferraris¹, D. Caruso², M. Francone², A. Laghi³; ¹Torino/IT, ²Rome/IT, ³Pieve Emanuele/IT
Purpose: To assess the accuracy of RECIST 1.1 in evaluating treatment response after neoadjuvant chemotherapy, with or without immunotherapy, in patients with stage IIIA–IIIB non-small cell lung cancer (NSCLC) undergoing surgical resection, and to highlight its limitations in reflecting true pathological outcomes.
Methods or Background: In a prospective pilot study, 33 patients with stage IIIA–IIIB NSCLC received neoadjuvant chemotherapy prior to surgery; 11 also received pembrolizumab. All patients underwent baseline and post-treatment contrast-enhanced CT scans, evaluated using RECIST 1.1 or iRECIST by two radiologists in consensus. Surgical specimens were analyzed histopathologically to determine complete or partial response. Radiological and pathological responses were then compared.
Results or Findings: Of the 33 patients, 12 achieved complete pathological response (pCR). None of these patients were classified as complete responders by RECIST 1.1 on post-treatment imaging. The remaining 21 patients showed partial pathological response, with residual viable tumor ranging from 10% to 80%. Only 9 of these (43%) met RECIST criteria for partial response, while the others were categorized as stable disease despite substantial histological regression. This discordance was more pronounced in patients treated with combined chemo-immunotherapy, suggesting that lesion size alone may fail to capture treatment-induced changes such as immune-related remodeling or necrosis.
Conclusion: RECIST 1.1 appears insufficiently sensitive to capture complete or substantial tumor response after neoadjuvant therapy in stage IIIA–IIIB NSCLC. The discrepancy is most evident with immunotherapy, which may induce changes not reflected by lesion size. As a result, radiological assessments often underestimate true pathological response. In locally advanced NSCLC, alternative or complementary imaging biomarkers are needed to better predict pathological outcomes and guide clinical decisions, particularly when chemotherapy is combined with immunotherapy.
Limitations: The limitations are the small sample size and single-center design
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Written informed consent was acquired for all patients and Istitutional Review Board approval was obtained.

Author Block: X. Jiang, J. Wang; Beijing/CN
Purpose: To test whether integrating preoperative contrast-enhanced CT (CECT) habitat radiomics with postoperative minimal residual disease (MRD) enables non-invasive, binary prediction of early recurrence after R0 resection of NSCLC.
Methods or Background: In this single-center prospective study (n=119), tumors on preoperative CECT were manually segmented; intratumoral habitats were derived by k-means clustering. Radiomics features (whole-tumor and habitat) underwent z-score normalization, variance/collinearity filtering, optional univariate screening, and L1-penalized logistic selection to form a logistic radiomics score (RadScore). Clinical screening identified MRD (first postoperative result, ±) and maximum diameter as independent predictors. Three classifiers were trained with five-fold internal CV on a stratified 70/30 split and tested on the hold-out set: radiomics-LR (RadScore), clinical (MRD+size), and combined-LR (RadScore+MRD+size). Performance was assessed by AUC, accuracy, sensitivity, specificity, calibration (Brier; intercept/slope), decision-curve analysis (DCA), and pairwise DeLong tests.
Results or Findings: On the test set, the combined-LR model showed the highest discrimination (AUC 0.867), exceeding radiomics-LR (0.847), clinical (0.786), and calibrated radiomics-SVM (0.773). Calibration favored the combined model (Brier 0.179; intercept −1.66; slope 0.95). DCA demonstrated greater net benefit for the combined model across thresholds ~0.05–0.60. DeLong comparisons were concordant with AUC ranking.
Conclusion: CECT-based habitat radiomics complements postoperative MRD by capturing spatial heterogeneity, enabling accurate, non-invasive prediction of early recurrence after NSCLC resection and supporting risk-adapted postoperative management.
Limitations: Single-center cohort and modest event counts may limit generalizability; external, multi-institutional validation is warranted.
Funding for this study: Not Applicable
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics Committee of Cancer Hospital, Chinese Academy of Medical Sciences

Author Block: F. Palmeri¹, V. Solimene², A. Del Gaudio², M. Zerunian², F. Di Gregorio², A. Ferraris¹, D. Caruso², M. Francone², A. Laghi³; ¹Torino/IT, ²Roma/IT, ³Pieve Emanuele/IT
Purpose: To develop and validate machine learning (ML) models based on radiomic features extracted from CT imaging to differentiate between high (PD-L1 ≥50%) and low (PD-L1 <50%) expression in non-small cell lung cancer (NSCLC). The aim was to explore whether imaging biomarkers could provide a noninvasive alternative to histological profiling for guiding immunotherapy decisions.
Methods or Background: This retrospective study included 210 patients with histologically confirmed NSCLC and pre-treatment contrast-enhanced chest CT. Among them, 46 (21.9%) showed high PD-L1 expression, while 164 (78.1%) showed low expression. Tumors were manually segmented, and radiomic features were extracted following Image Biomarker Standardization Initiative guidelines. Five ML models were built for binary classification (“PD-L1 ≥50%” vs. “PD-L1 <50%”), using Random Forest, Support Vector Machine, K-Nearest Neighbors, Multi-Layer Perceptron, and Logistic Regression classifiers. Histological PD-L1 served as the reference standard. Statistical significance was set at P < .05.
Results or Findings: The Multi-Layer Perceptron model achieved the best performance, with a ROC-AUC of 0.61 (95% CI: 0.57–0.66, P< .005) and an accuracy of 64% (95% CI: 60–68%). Sensitivity reached 70% (95% CI: 65–75%) and PPV 82% (95% CI: 80–84%), while specificity and NPV were lower at 44% (95% CI: 36–52%) and 29% (95% CI: 24–35%), respectively. The F1 score was 75% (95% CI: 71–78%).
Conclusion: Radiomic analysis of pre-treatment CT images using a Multi-Layer Perceptron classifier showed potential in distinguishing PD-L1 expression levels in NSCLC, with good sensitivity, PPV, and overall F1 score. However, the model's limited specificity and NPV suggest it may be more effective in identifying patients likely to express high PD-L1, rather than excluding them. Radiomic-based ML models could aid noninvasive prediction of PD-L1, supporting immunotherapy selection when biopsy is inconclusive or risky.
Limitations: The limitations are the retrospective single-center design.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Written informed consent was acquired for all patients and Istitutional Review Board approval was obtained.

Author Block: Z. Lin¹, M. Zuo², J. WANG², Y. Tan¹, Y. Wang³, X. Yu³; ¹Hubei/CN, ²Nanchang/CN, ³Shanghai/CN
Purpose: To develop and validate a radiomics model based on multi-parametric Dual-layer CT (DLCT) images for non-invasive differentiation of benign and malignant solid solitary pulmonary nodules (SSPNs).
Methods or Background: This retrospective study included 159 patients with pathologically confirmed SSPN (64 benign, 95 malignant) who underwent DLCT-enhanced scanning, randomly divided into training (n=111) and test (n=48) cohorts at a 7:3 ratio. Radiomic features were extracted from seven venous-phase image series: conventional images (CI), iodine density (ID) maps, effective atomic number (Zeff) maps, electron density (ED) maps, virtual monochromatic images (VMI) at 40 keV and 100 keV, and virtual non-contrast (VNC) images. Logistic regression models were built for each series, and features were integrated to construct a multi-DLCT model.
Results or Findings: The CI, ID, Zeff, ED, VMI 40 keV, VMI 100 keV, and VNC models achieved AUCs of 0.777, 0.830, 0.812, 0.783, 0.792, 0.797, and 0.790 in the training cohort, and 0.774, 0.789, 0.741, 0.804, 0.737, 0.741, and 0.735 in the test cohort. The multi-DLCT model demonstrated the highest discriminatory performance, achieving AUCs of 0.832 in training and 0.863 in testing, and outperformed the CI model (AUC = 0.777 training, 0.774 test), although differences were not statistically significant (Delong P = 0.101 and 0.232, respectively). Positive integrated discrimination improvement (IDI) and net reclassification improvement (NRI) values indicated improved discrimination, and decision curve analysis showed greater net clinical benefit for the multi-DLCT model.
Conclusion: The radiomics model based on multi-parametric DLCT improves differentiation of benign and malignant SSPNs and provides higher clinical utility than conventional imaging.
Limitations: Single-center study with a limited sample size.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of the Second Affiliated Hospital of Nanchang University.