CT biomarkers assess immunotherapy response and surgical planning in lung cancer

Computed tomography can help clinicians interpret treatment response and plan surgery after neoadjuvant immunotherapy for resectable non-small cell lung cancer, but it cannot reliably show how much viable tumor remains. A 2026 mini review concludes that conventional CT is established for describing anatomy, lymph nodes, complications and operative complexity, while dual-energy measurements, radiomics and deep-learning models remain investigational tools for predicting pathological response. The review says these approaches should complement clinical, pathological and molecular information rather than be used to withhold surgery from a patient whose cancer remains resectable.

Immunotherapy has changed the relationship between a lung lesion that persists on a post-treatment scan and the biological response within it. Immune-cell infiltration, edema, necrosis, granulomatous reactions and reparative fibrosis can maintain or increase apparent tumor dimensions even after major treatment effects, while small nests of viable cancer may remain within an apparently fibrotic tumor bed. Consequently, a radiographically stable mass can represent complete pathological response, major pathological response or substantial residual disease. By contrast, a modest reduction in diameter may accompany clinically important residual tumor. Imaging and pathology therefore measure different aspects of response, and their disagreement is a biological limitation rather than necessarily a technical failure.

The most established role of contrast-enhanced chest CT is presurgical intrathoracic restaging and assessment of operative anatomy; it does not replace appropriate distant or brain staging. RECIST 1.1 can describe changes in tumor size, and iRECIST can accommodate confirmation of initially unconfirmed progression, usually after four to eight weeks, but neither is a validated surrogate for major pathological response or pathological complete response. Major pathological response generally means no more than 10% residual viable tumor in the treated primary-tumor bed, whereas pathological complete response requires no viable tumor in the primary tumor and sampled regional nodes. These endpoints are determined after surgery and vary with nodal assessment, tumor-bed sampling and immune-related pathological criteria, so a reduction in lesion diameter cannot be assumed to predict them reliably.

Serial interpretation offers more information than an isolated presurgical scan. Radiologists should compare pretreatment and post-treatment images side by side, document tumor regression alongside new inflammatory changes and state whether an abnormality is more likely to represent viable tumor, treatment effect or an indeterminate mixture. Candidate features include percentage change in longest diameter, volumetric reduction, attenuation change, cavitation, loss of a solid component, restoration of an air bronchogram and alteration of tumor contact with vessels or bronchi. The report should also describe pleural contact, chest-wall invasion, fissural extension, vascular encasement, bronchial narrowing and nodal distribution. Comparable contrast phase, slice thickness, reconstruction kernel, respiratory conditions and segmentation are necessary for dependable serial measurements, and the intervals from the last treatment to CT and from CT to surgery should be recorded.

Dual-energy CT may add quantitative information beyond conventional morphology. Measurements such as iodine concentration, normalized iodine concentration, effective atomic number and virtual monoenergetic reconstructions can reflect perfusion and tissue characteristics. The reviewed evidence remains exploratory because inflammation, contrast timing, iodine normalization and scanner implementation can influence these measurements, and no universal post-immunotherapy thresholds have been established. A study combining single-source dual-energy CT with tumor-microenvironment features offered exploratory evidence about disagreement between RECIST 1.1 response and pathological remission, but it did not establish an imaging substitute for pathology. Dual-energy CT can therefore help characterize treatment-associated tissue changes, but it is not yet a validated stand-alone method for determining residual cancer or selecting a particular operation.

Radiomics and deep learning take a different approach by converting CT images into quantitative descriptors of shape, texture, intensity distribution and temporal change. Pretreatment models estimate the probability of subsequent pathological response, post-treatment models attempt to recognize the residual tumor bed, and delta-radiomics measures changes in features between scans. In a four-centre retrospective study of 274 patients receiving neoadjuvant chemoimmunotherapy, a combined clinical and deep-learning model achieved an external area under the receiver operating characteristic curve of 0.75, with a 95% confidence interval of 0.62–0.87, for predicting major pathological response. In another retrospective study of 263 patients with stage II–III disease, the combined model’s AUC declined from 0.89 in development to 0.78 in external testing; its external advantage over delta-radiomics alone was not statistically significant. These results support task-specific evaluation rather than a general claim that temporal models are superior.

Preliminary prospective testing has produced a reader-support signal without showing that artificial intelligence is ready for autonomous decisions. In the 50-patient prospective cohort of the multicentre NeoPred study, a dual-time-point CT model achieved an AUC of 0.760, with a 95% confidence interval of 0.628–0.891, for major pathological response. Reported expert performance was 0.720, while AI-assisted experts achieved 0.829. The small sample and absence of a prospective management-impact trial do not establish deployment readiness. Retrospective surgical cohorts also introduce selection and verification bias because patients who develop progression, severe toxicity or unresectable disease may never reach surgery. Stage, histology, treatment regimen, tumour size, centre, scanner vendor, reconstruction and segmentation can confound performance, while random internal splits do not demonstrate that a model will transport to a new hospital or clinical setting.

Lymph-node restaging is a particularly high-risk application because an immune response can make nodes enlarge. In an ad hoc NEOSTAR analysis, nodal immune flare occurred in 7 of 44 patients treated with immune checkpoint inhibitors, compared with none of 28 patients in a historical chemotherapy cohort. The small, non-equivalent groups do not establish a universal incidence, but they show why a new or enlarged mediastinal node cannot automatically be labeled as progression. When the result would preclude curative surgery or redirect treatment, tissue confirmation through an appropriate route, such as endobronchial ultrasound-guided sampling or mediastinoscopy, should be considered. Radiographic nodal regression likewise does not eliminate the need for systematic nodal evaluation during surgery.

CT can also reveal treatment-related anatomy that complicates an operation. Hilar inflammation and fibrosis may obscure tissue planes around pulmonary arteries, veins and bronchi, suggesting difficult dissection without proving tumour invasion. In the primary NEOSTAR surgical analysis, 42 of 44 patients underwent resection; 22 of the 42 resections were considered challenging, and three pneumonectomies were performed. Grade 3 or higher complications occurred in four patients in each of the two small sequential treatment cohorts. These findings support careful reporting of tumour relationships with the pulmonary artery, veins, lobar and main bronchi, fissures, chest wall, mediastinum and adjacent lobes, but they do not constitute validation of a CT-based operative-complexity score. Presurgical imaging can help the multidisciplinary team prepare for technically demanding dissection and the possible need to change the planned operation.

Current practice therefore centers on structured interpretation linked to specific clinical actions. Radiologists, thoracic surgeons, medical oncologists, pulmonologists and pathologists can use CT to identify suspicious targets for invasive restaging, assess complications, define operative anatomy and estimate the likelihood of pathological response while acknowledging uncertainty. A persistent mass alone should not justify omitting or postponing surgery when disease remains resectable and progression has not been verified, and no imaging model predicting complete response is sufficiently mature to support nonoperative management. Future research should predefine the clinical decision each biomarker is intended to improve, harmonize acquisition and scan timing, include patients who do not reach surgery, test reproducibility across institutions and scanners, and report calibration, decision-curve performance, subgroup results and failure cases alongside discrimination. Prospective validation is needed to determine whether model-assisted imaging safely changes management. Until then, CT’s strongest contribution is to make uncertainty explicit and connect imaging to tissue confirmation, surgical preparation and multidisciplinary review, not to replace pathology or decisive clinical judgement.

Subject of Research: CT assessment of treatment response and surgical planning in resectable non-small cell lung cancer

Article Title: CT imaging biomarkers after neoadjuvant immunotherapy in resectable non-small cell lung cancer: from response assessment to surgical decision-making

Article References: Li, Y., & Wen, L. (2026). CT imaging biomarkers after neoadjuvant immunotherapy in resectable non-small cell lung cancer: from response assessment to surgical decision-making. Frontiers in Medicine, 13, Article 1949375. https://doi.org/10.3389/fmed.2026.1949375

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1949375

Keywords: non-small cell lung cancer, neoadjuvant immunotherapy, computed tomography, pathological response, radiomics, deep learning, dual-energy CT, surgical decision-making, lymph-node restaging, operative anatomy, imaging, biomarkers