Quantitative PCCT spectral parameters for noninvasive prediction of EGFR status and its subtypes in lung adenocarcinoma

Abstract

Objective

To investigate the non-invasive predictive value of quantitative spectral parameters from photon-counting computed tomography (PCCT) for epidermal growth factor receptor (EGFR) mutation status and its predominant subtypes (19Del and L858R) in patients with lung adenocarcinoma.

Methods

A total of 72 patients with pathologically confirmed lung adenocarcinoma who underwent pretreatment PCCT were retrospectively enrolled. Arterial and venous CT attenuation values at 40 keV, 70 keV, and 100 keV (A/V-40 keV, A/V-70 keV, A/V-100 keV) were measured on virtual monoenergetic images, while arterial/venous iodine concentration (IC) and dual-energy index (DEI) of lesions were measured on iodine maps and spectral post-processing (SPP) images, respectively. Normalized iodine concentration (NIC) and spectral curve slope (λHU) were further calculated. Receiver operating characteristic (ROC) curve analysis and binary logistic regression were performed to evaluate the predictive performance and independent predictive value of PCCT parameters for discriminating EGFR-mutant vs. wild-type tumors, as well as 19Del vs. L858R subtypes.

Results

Of 72 patients, 37 (51.4%) harbored EGFR mutations, which correlated with female sex, never-smoking, reduced NSE, and lower monocyte count. The EGFR-mutant group showed significantly higher A-70 keV, A-100 keV, A-DEI, V-40 keV, V-70 keV, V-100 keV, V-λHU and V-DEI (all P < 0.05). Logistic regression identified female sex (P = 0.019, OR = 5.714, 95% CI: 1.336–24.442) and A-100 keV (P = 0.044, OR = 0.543, 95% CI: 0.300-0.983) as independent predictors. ROC analysis with bootstrap validation yielded: clinical model AUC 0.767 (optimism-corrected 0.762); PCCT model AUC 0.713 (optimism-corrected 0.707); combined model AUC 0.771 (optimism-corrected 0.768) (all P < 0.001). Between 19Del (n = 15) and L858R (n = 17) subgroups, A-40 keV and A-DEI showed significant discriminative performance in ROC analysis, with AUCs of 0.694 and 0.704 (optimism-corrected AUCs: 0.688 and 0.697, respectively).

Conclusions

Quantitative PCCT spectral parameters enable non-invasive prediction of EGFR mutation status and preliminary differentiation between 19Del and L858R subtypes in lung adenocarcinoma, though subtype-related findings require validation in larger cohorts. Female sex and A-100 keV are independent predictive factors. The combined model yielded a numerically higher AUC without statistically significant superiority, and PCCT parameters may provide auxiliary imaging evidence to inform individualized targeted therapy decisions in lung adenocarcinoma.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40644-026-01122-9.

Introduction

Lung cancer, the leading cause of cancer-related mortality globally, accounts for 11–12% of all malignancies, with non-small cell lung cancer (NSCLC) representing the predominant pathological subtype [1]. With the rapid development of targeted therapy, epidermal growth factor receptor (EGFR) mutation testing has become central to personalized lung adenocarcinoma care, as tyrosine kinase inhibitors (TKIs) significantly prolong survival and improve quality of life in EGFR-mutant patients [2]. Currently, TKIs are recommended as first-line therapy for EGFR-mutant stage IV NSCLC, and EGFR mutation status is a mandatory prerequisite for initiating TKIs treatment [3, 4].

However, routine EGFR mutation detection in clinical practice is mainly dependent on invasive procedures such as endoscopic and percutaneous biopsy, which have notable limitations: they carry risks of complications like bleeding and infection, are poorly tolerated by frail advanced patients, and biopsy samples only reflect local tumor genetics, failing to capture tumor heterogeneity. Inadequate sampling may also result in false-negative or false-positive results, compromising treatment decision accuracy [5]. Thus, developing a non-invasive, convenient and accurate EGFR mutation detection method is of great clinical significance for optimizing TKI therapy and improving outcomes in lung adenocarcinoma patients.

Currently, available non-invasive methods for EGFR mutation detection have notable limitations that hinder their clinical utility in precision care. Liquid biopsy using peripheral blood, while non-invasive, has low sensitivity, long turnaround times and high costs, limiting its widespread use [6]. Although imaging parameters and radiomic features from conventional CT and ¹⁸F-fluorodeoxyglucose (¹⁸FDG) positron emission tomography (PET)/computed tomography (CT) allow preliminary prediction of EGFR mutation status in lung adenocarcinoma, their overall predictive performance is limited due to the lack of precise quantitative indicators [7–9]. Conventional spectral CT offers multi-parameter quantitative analysis and has shown value in EGFR mutation prediction [10–12], but is limited by technical constraints of traditional energy-integrated detector CT (EID-CT), with insufficient spatial resolution, noise control and quantitative accuracy to meet the demands of precise clinical prediction.

As a new-generation advanced oncologic imaging technique, photon-counting CT (PCCT) addresses technical limitations of conventional EID-CT and provides new opportunities for precision diagnosis and treatment of tumors. Compared with conventional EID-CT, PCCT directly converts X-ray photons into electrical signals via electron-hole-pair generation. This allows for precise identification and measurement of individual photon energies and mitigates cross-talk and segmentation effects among detector elements. This results in higher spatial resolution, lower image noise, and improved tissue contrast and signal-to-noise ratio (SNR) [13]. Meanwhile, PCCT enables efficient and accurate spectral data acquisition and allows multi-dimensional quantitative assessment of tumor tissues. Compared with conventional EID-CT, PCCT has prominent advantages in multi-parameter quantitative analysis, energy resolution and quantitative accuracy, thereby providing more robust technical support for the noninvasive prediction of genetic status.

Based on the above background and the unique advantages of PCCT, this study focused on the value of PCCT spectral parameters, aiming to systematically investigate their feasibility and efficacy in predicting EGFR mutations in lung adenocarcinoma. It is expected to provide new ideas and methods for noninvasive clinical detection of EGFR mutations, offer more precise imaging evidence for personalized targeted therapy, and promote the advancement of precision diagnosis and treatment for lung adenocarcinoma.

Materials and methods

Patients

This study was approved by the Institutional Review Board of the Cancer Hospital Affiliated to Shandong First Medical University according to the Helsinki Declaration, and the requirement to obtain informed consent from each participant was waived due to the retrospective study. And this study was conducted at a single center.

We retrospectively reviewed the radiology database and collected a total of 198 patients who were suspected with pulmonary tumors and received pulmonary scanning with PCCT at our institution between May 2025 and March 2026. The inclusion criteria were as follows: [1] Patients diagnosed with lung adenocarcinoma by surgical resection or biopsy; [2] PCCT was performed within 2 weeks before surgery or biopsy, with diagnostic-quality images available for all patients; [3] Age ≥ 18 years; [4] Complete clinical and pathological data were available. The exclusion criteria included: [1] Histologically confirmed squamous cell carcinoma, SCLC, or combined small cell lung cancer; [2] Patients who had received any preoperative antitumor treatment, including chemotherapy, radiotherapy, targeted therapy, or immunotherapy; [3] Images suffering from severe artifacts, motion artifacts or excessive slice thickness were excluded due to failure of tumor segmentation and feature extraction; [4] Comorbidity with other malignant tumors. Finally, 72 lung adenocarcinoma patients were entered into the present study. The research flowchart is presented in Fig. 1.

Fig. 1.

Flowchart for patient enrollment

Pretreatment clinical data — including age, gender, TNM stage, smoking history, serum tumor markers, and peripheral hematological parameters — were extracted from medical records. The tumor markers and their reference ranges were: carcinoembryonic antigen (CEA, 0–5 ng/mL), neuron-specific enolase (NSE, 0–17 ng/mL), cytokeratin 19 fragment (CYFRA21-1, 0–3.3 ng/mL), pro-gastrin-releasing peptide (ProGRP, 0–65.7 pg/mL), and squamous cell carcinoma–associated antigen (SCC-Ag, 0–2.7 ng/mL). Hematological parameters included white blood cell count (WBC, 3.5–9.5 × 10⁹/L), neutrophil count (Neu, 1.8–6.3 × 10⁹/L), monocyte count (MO, 0.10–0.60 × 10⁹/L), absolute lymphocyte count (ALC, 1.1–3.2 × 10⁹/L), red blood cell count (RBC, 3.8–5.1 × 10¹²/L), platelet count (PLT, 125–350 × 10⁹/L), hemoglobin (HGB, 130–175 g/L), albumin (ALB, 40–55 g/L), prealbumin (PAB, 0.25–0.40 g/L), and hematocrit (Hct, 40–50%). Derived ratios were calculated as follows: neutrophil-to-lymphocyte ratio (NLR) = Neu/ALC, platelet-to-lymphocyte ratio (PLR) = PLT/ALC, and lymphocyte-to-monocyte ratio (LMR) = ALC/MO.

CT examination

All participants underwent contrast-enhanced chest CT on a clinical dual-source photon-counting CT system (NAEOTOM Alpha, Siemens Healthineers, Forchheim, Germany). Scanning was conducted at 140 kVp with an image quality (IQ) setting of 80, a collimation of 144 × 0.4 mm, a pitch of 0.8, and a gantry rotation time of 0.5 s. Spectral post-processing (SPP) images and iodine maps were reconstructed with 1-mm slice thickness, 0.7-mm increment, quantum iterative reconstruction (QIR) at strength level 3, and a Qr40 kernel, then transferred to a dedicated workstation (syngo.Via VB60; Siemens Healthineers, Forchheim, Germany) for subsequent analysis.

Iodine-based contrast medium (370 mgI/mL; Bayer Healthcare) was administered intravenously through a peripheral vein at a weight-adjusted dose of 1.2 mL/kg and an injection rate of 2.5 mL/s. The bolus-tracking technique was employed, with a region of interest (ROI) positioned in the thoracic aorta at the cardiac level. Following attainment of the predefined attenuation threshold of 140 HU within the ROI, arterial phase acquisition commenced 25 s later, and portal venous phase images were subsequently obtained 50 s thereafter.

Imaging analysis

All image data were processed and analyzed using the aforementioned workstation. Tumor ROIs were independently delineated by two radiologists with 8 and 5 years of thoracic imaging experience, respectively, who were fully blinded to all clinical and pathological data. Any disagreements were resolved by mutual consensus following open discussion. Necrotic, cystic, and calcified areas were avoided during ROI1 delineation. Within the same imaging phase, ROI1 was placed on the largest axial slice of the lesion’s largest diameter as well as on the two adjacent slices above and below (three consecutive slices in total). The mean value of these three slices was used as the final result. If the lesion was heterogeneous in density, the slice with the most solid component was selected for measurement. Additionally, a circular ROI measuring 5–10 mm² (ROI2) was placed within the descending aorta or subclavian artery at the corresponding slice level, and its iodine concentration was measured on iodine maps for normalization. Iodine concentrations of ROI1 in the arterial and venous phases (A-IC/V-IC) were also measured on iodine maps.

Virtual monoenergetic images at 40 keV, 70 keV, and 100 keV were reconstructed for both arterial and venous phases, and the corresponding CT attenuation values (A/V-40 keV, A/V-70 keV, and A/V-100 keV) were measured within the primary tumor ROI (ROI1). The spectral HU slope (λHU) was calculated as λHU = (CT40 keV − CT100 keV) / (100 − 40), where CT40 keV and CT100 keV correspond to the A/V-40 keV and A/V-100 keV values, respectively, as previously reported. The normalized iodine concentration (NIC) was defined as the ratio of lesion iodine concentration to aortic iodine concentration at the corresponding phase, calculated as A-NIC = A-IC / A-ICaorta and V-NIC = V-IC / V-ICaorta, where A-ICaorta and V-ICaorta denote the aortic iodine concentrations measured from ROI2 in the arterial and venous phases, respectively. Effective atomic number (Zeff) and dual-energy index (DEI) were measured on spectral post-processing (SPP) images loaded on the workstation for each tumor ROI. The arterial enhancement fraction (AEF) was calculated as AEF = (A-IC − V-IC) / A-IC, where A-IC and V-IC denote lesion iodine concentrations acquired from arterial and venous iodine maps, respectively. PCCT-derived extracellular volume (ECV) was calculated using the formula ECV = (1 − hematocrit [Hct]) × (V-IC / V-ICaorta), where Hct refers to the volumetric fraction of red blood cells in whole blood, and V-ICaorta represents the venous-phase aortic iodine concentration measured via ROI2.

EGFR mutation analysis

EGFR mutation analysis was performed on tumor histologic specimens. Using the Human EGFR Gene Mutations Detection Kit (Amoy Diagnostics, Xiamen, China) based on real-time quantitative polymerase chain reaction (qPCR) combined with the amplification refractory mutation system (ARMS), exons 18–21 of the EGFR gene were amplified. Detected mutations included exon 19 deletion (19del); exon 21 L858R and L861Q; exon 20 T790M, insertion mutation (20ins), and S768I; and exon 18 G719X. According to the kit’s interpretation criteria, the presence of any of the above mutations was defined as EGFR-mutant, whereas the absence of these mutations was defined as wild-type.

Statistical analysis

Statistical analysis was performed using SPSS software (Windows version 27.0; SPSS Inc., Chicago, IL, USA) and R software (version 4.4.1, R Foundation for Statistical Computing, Vienna, Austria). Receiver operating characteristic (ROC) curve analysis was conducted using MedCalc statistical software (version 22.0, MedCalc Software Ltd, Ostend, Belgium). Continuous data were assessed for normality using the one‑sample Shapiro‑Wilk test. Measurement data with a normal distribution were presented as the mean ± standard deviation, and comparisons between two groups were performed using the independent-samples t-test. Non-normally distributed data were expressed as the median (interquartile range, IQR), and comparisons between two groups were performed using the Mann‑Whitney U test. Categorical data were reported as the number of cases, and the chi‑square test was applied for comparison. Intraclass correlation coefficients (ICCs) were calculated to evaluate interobserver reproducibility of all PCCT spectral parameters measured independently by two masked radiologists. For parameters with statistically significant differences, ROC curves were plotted to examine the predictive efficacy of each quantitative parameter; combined ROC analysis and the area under the curve (AUC) were applied to assess the predictive performance of the combined parameters. Bootstrap internal validation with 1,000 resampling iterations was conducted for all ROC analyses. Optimism-corrected AUCs and 95% BCa bootstrap confidence intervals were calculated to adjust for overfitting bias and verify out-of-sample predictive stability. Binary logistic regression analysis was performed to explore the associations between clinical variables (gender, smoking history, NSE, MO), PCCT quantitative spectral parameters and EGFR mutation status. Categorical variables were analyzed with original dummy coding, and all continuous variables were Z-standardized before regression. Odds ratios for continuous variables are presented per one-standard-deviation increment. To avoid multicollinearity, collinearity diagnostics were performed via linear regression to calculate variance inflation factor (VIF) and tolerance; VIF > 10 indicated severe multicollinearity, and relevant variables were excluded for model stability. All tests were two-sided, and the P value smaller than 0.05 was considered statistically significant.

Results

Patient clinical and pathologic characteristics

Of the 72 patients, 39 were male and 33 were female, with a mean age of 64.24 ± 1.03 years. Of all patients, 37 patients were mutated EGFR and 35 were EGFR wild-type (Fig. 2). With the exception of gender (P < 0.001), smoking history (P = 0.009), NSE (P = 0.010) and MO (P = 0.042), there were no significant differences in age, clinical stage, or other laboratory examinations among groups with different EGFR status (all P > 0.05). The clinical and pathological characteristics of these lung adenocarcinoma patients with different EGFR status are illustrated in Table 1.

Fig. 2.

PCCT parameters and pathological images in lung adenocarcinoma patients with distinct EGFR mutation status. (A–D) Patient 1 with adenocarcinoma in the right upper lobe (EGFR L858R mutation): 40 keV virtual monoenergetic image, fusion image, effective atomic number (Zeff) image, and pathological image. (E–H) Patient 2 with adenocarcinoma in the right lower lobe (EGFR 19 Del mutation): 40 keV virtual monoenergetic image, fusion image, effective atomic number (Zeff) image, and pathological image

Table 1.

Clinical and pathologic characteristics of patients(n = 72)

Characteristics		Mutated EGFR group	Wild-type EGFR group	χ²/Z/t	P
No. of patients		37 (51.4%)	35 (48.6%)
Age	Mean ± SD	63.86 ± 6.33	64.75 ± 8.98	-1.629	0.109
Gender	Male	12	27	13.758	<0.001
	Female	25	8
Stage	I-II	6	7	0.094	0.717
	III-IV	30	28
Smoking history	Smoking	9	20	7.721	0.009
	Never smoking	24	14
Laboratory parameters	CEA (ng/mL)	38.95 (4.17, 271.50)	13.95 (4.41, 89.50)	-0.840	0.401
	NSE (ng/mL)	16.39 ± 3.66	22.14 ± 8.37	-3.216	0.010
	CYFRA21-1 (ng/mL)	6.31 (2.43,9.60)	6.09 (2.58, 15.33)	-0.862	0.389
	ProGRP(pg/mL)	48.91 ± 12.58	45.30 ± 21.07	-0.172	0.864
	SCC-Ag (ng/mL)	0.69 (0.53,1.04)	1.08 (0.72,1.86)	-0.878	0.380
	WBC (×109/L)	6.56 ± 1.56	7.41 ± 2.35	-1.704	0.094
	ALC(×109/L)	1.41 (1.07,1.96)	1.66 (1.35,2.00)	-1.006	0.314
	NEU(×109/L)	4.31 (3.42,5.09)	4.60 (3.71,6.41)	-0.811	0.417
	MO(×109/L)	0.42 ± 0.14	0.51 ± 0.20	-2.075	0.042
	RBC (×1012/L)	4.65 (4.41,4.90)	4.59 (4.36,5.04)	-0.650	0.516
	HGB (g/L)	138.00 (125.00,144.00)	139.00 (129.00,154.00)	-1.755	0.079
	PAB (g/L)	0.24(0.19,0.27)	0.24 (0.20,0.27)	-0.194	0.846
	ALB (g/L)	41.80 (39.40,43.50)	42.00 (39.70,43.70)	-0.435	0.664
	Hct (%)	41.03 ± 3.71	43.14 ± 4.58	-1.931	0.060
	PLT(×109/L)	283.43 ± 65.98	263.29 ± 59.13	1.560	0.123
	NLR	2.96 (1.90,3.85)	2.72 (2.21,3.78)	-0.069	0.945
	PLR	176.62 (140.65,236.92)	148.80 (128.08,211.61)	-1.806	0.071
	LMR	3.44 (2.76,4.74)	3.38 (2.60,5.13)	-0.661	0.508

CEA= carcinoembryonic antigen; NSE= neuron-specific enolase; CYFRA21-1 = cytokeratin 19 fragment; ProGRP = pro-gastrin-releasing peptide; SCC-Ag=squamous cell carcinoma–associated antigen; WBC= white blood cell count; Neu=neutrophil count; MO=monocyte count; ALC=absolute lymphocyte count; RBC = red blood cell count; PLT=platelet count; HGB=hemoglobin; ALB=albumin; PAB=prealbumin; Hct= hematocrit; NLR=neutrophil-to-lymphocyte ratio; PLR=platelet-to-lymphocyte ratio; LMR=lymphocyte-to-monocyte ratio

Note: P < 0.05. Clinicopathological characteristics were compared by chi-square test, Mann‑Whitney U test, or t-test across different EGFR gene status groups. Smoking history data were unavailable for five patients (four in the EGFR-mutant group and one in the wild-type group); some patients also lacked records of laboratory indicators such as CEA. All cases with missing data for these indicators were excluded from the corresponding statistical analyses

Most EGFR mutations associated with lung adenocarcinoma are in-frame deletions in exon 19 (19del) and a leucine-to-arginine point mutation in exon 21 (L858R) [14]. Among the 37 patients with EGFR mutations, 15 patients harbored exon 19 deletion and 17 had exon 21 L858R mutation.

Comparison of quantitative PCCT parameters in patients with different EGFR status

The interobserver ICC for PCCT spectral parameters was 0.87 demonstrating excellent reproducibility between the two radiologists. Except for arterial phase parameters A‑70 keV (P = 0.012), A‑100 keV (P = 0.005), A‑DEI (P = 0.039), and venous phase parameters V‑40 keV (P = 0.023), V‑70 keV (P = 0.018), V‑100 keV (P = 0.029), V‑λHU (P = 0.030) and V‑DEI (P = 0.044), no significant differences were found in the remaining PCCT spectral parameters, as shown in Table 2.

Table 2.

Comparison of quantitative PCCT spectral parameters in lung adenocarcinoma patients with different EGFR status

Characteristics	Mutated EGFR group	Wild-type EGFR group	t/Z	P
A-IC	0.75 ± 0.38	0.60 ± 0.33	1.705	0.093
A-NIC	0.09 (0.07, 0.13)	0.08 (0.05, 0.11)	-1.519	0.129
A-40 keV	102.57 ± 32.01	89.74 ± 29.99	1.743	0.086
A-70 keV	59.84 ± 11.45	53.29 ± 10.97	2.458	0.012
A-100 keV	48.89 ± 7.17	44.08 ± 6.96	2.863	0.005
A-λHU	0.91 ± 0.45	0.77 ± 0.43	1.319	0.134
A-Zeff	7.80 (7.70, 8.00)	7.70 (7.60, 7.90)	-0.191	0.849
A-DEI	0.009 (0.005, 0.010)	0.006 (0.003, 0.009)	-2.064	0.039
AEF	0.53 (0.36,0.67)	0.50 (0.38, 0.68)	-0.045	0.964
V-IC	1.40 (1.08,1.67)	1.03 (0.80, 1.43)	1.906	0.062
V-NIC	0.32 (0.29,0.46)	0.29 (0.21, 0.39)	-1.302	0.193
V-40 keV	153.34 ± 40.32	132.98 ± 36.66	2.225	0.023
V-70 keV	74.19 ± 13.55	67.07 ± 11.89	2.352	0.018
V-100 keV	54.16 ± 7.57	50.38 ± 6.89	2.197	0.029
V-λHU	1.67 ± 0.58	1.39 ± 0.54	2.117	0.030
V-Zeff	8.18 ± 0.27	8.09 ± 0.26	1.456	0.150
V-DEI	0.014 (0.009,0.016)	0.010 (0.009, 0.014)	-1.908	0.044
ECV	0.21 (0.16,0.26)	0.18 (0.12, 0.22)	-2.676	0.057

A= arterial phase; AEF= arterial enhancement fraction; DEI= dual-energy index; ECV= extracellular volume; IC= iodine concentration; NIC= normalized iodine concentration; V= portal venous phase; Zeff= effective atomic number; λHU= slope of the spectral HU curve

P< 0.05. PCCT spectral parameters were compared between the EGFR-mutant and wild-type groups using the Mann–Whitney U test or t-test

Of the 15 patients with 19Del and 17 with L858R mutation, only A‑DEI (P = 0.049) exhibited a statistically significant difference, whereas all other PCCT spectral parameters showed no significant differences (all P > 0.05), as shown in Table S1.

Independent predictive value of quantitative PCCT parameters for EGFR mutation

Prior to regression analysis, all continuous variables were Z-standardized, while categorical variables retained original dummy coding. Univariate and multivariate binary logistic regression analyses were performed to identify independent predictors of EGFR mutation status. For all Z-standardized continuous predictors, odds ratios (ORs) reflect the change in log-odds per one standard deviation increase in the parameter.

Univariate binary logistic regression analysis identified the following clinical factors significantly associated with EGFR status: gender (P < 0.001, OR = 6.750, 95% CI: 2.362–19.290), smoking history (P = 0.011, OR = 4.286, 95% CI: 1.497–12.272), NSE (P = 0.017, OR = 2.063, 95% CI: 1.141–3.731), and MO (P = 0.041, OR = 1.707, 95% CI: 1.023–2.851). The Z score-normalized PCCT spectral parameters significantly associated with EGFR status were as follows: arterial phase parameters A-70 keV (P = 0.016, OR = 0.519, 95% CI: 0.304–0.887), A-100 keV (P = 0.009, OR = 0.472, 95% CI: 0.269–0.827), A-DEI (P = 0.042, OR = 0.514, 95% CI: 0.271–0.975); and venous phase parameters V-IC (P = 0.018, OR = 0.519, 95% CI: 0.302–0.894), V-40 keV (P = 0.029, OR = 0.558, 95% CI: 0.331–0.941), V-70 keV (P = 0.022, OR = 0.546, 95% CI: 0.325–0.917), V-100 keV (P = 0.033, OR = 0.577, 95% CI: 0.347–0.958), V-λHU (P = 0.036, OR = 0.572, 95% CI: 0.340–0.963). In addition, ECV was also significantly associated with EGFR status (P = 0.038, OR = 0.578, 95% CI: 0.344–0.969).

Collinearity diagnostics revealed moderate collinearity between arterial phase A-70 keV and A-100 keV (VIF = 7.644 and 7.633, respectively), with a maximum condition index of 40.731 indicating strong collinearity primarily driven by the correlation between the two keV parameters. Although A-DEI showed no collinearity (VIF = 1.022), its effect estimate was unstable (P = 0.042, OR = 0.514, 95% CI: 0.271–0.975). To mitigate multicollinearity and ensure stable effect estimates, we selected A-100 keV as the representative arterial phase spectral parameter, as it demonstrated the most robust and statistically significant association with EGFR status (P = 0.009, OR = 0.472, 95% CI: 0.269–0.827). These collinearity diagnostics are presented in Table S2. Due to severe multicollinearity among venous-phase keV parameters (with a maximum VIF of 1578), V-λHU yielded the lowest VIF among venous-phase parameters; however, its VIF of 71.3 still exceeded the conventional threshold of 10. Compared with venous-phase 40 keV, 70 keV and 100 keV parameters, V-λHU was more representative. To avoid strong correlation between these physically homologous features, only V-λHU was retained as the representative iodine-related parameter, and other venous-phase keV parameters were excluded from the regression model to guarantee stable effect estimates, as shown in Table S3. Although extracellular volume (ECV) showed a marginally significant association with EGFR status in univariate analysis (P = 0.038, OR = 0.578, 95% CI: 0.344–0.969), the modest effect size and relatively wide confidence interval suggested limited statistical stability; therefore, ECV was not included in the multivariate analysis.

A multivariate binary logistic regression analysis incorporating clinical factors and Z‑score‑normalized PCCT spectral parameters demonstrated that gender (P = 0.019, OR = 5.714, 95%CI:1.336–24.442) and A‑100 keV (P = 0.044, OR = 0.543, 95%CI:0.300-0.983) were independent predictors of EGFR mutation status. Notably, V‑λHU entered the multivariate model but failed to reach statistical significance (P = 0.281, OR = 0.731, 95%CI:0.413–1.292). All logistic regression results calculated on Z-score-normalized variables are summarized in Table 3, while regression analyses using original non-normalized variables are presented in Table S4.

Table 3.

Univariate and multivariate binary logistic regression analyses for predicting EGFR mutation status in lung adenocarcinoma

Factor	Univariate analysis		Multivariate analysis
	OR (95% CI)	P	OR (95% CI)	P
Clinical factors
Gender	6.750 (2.362,19.290)	<0.001	5.714 (1.336,24.442)	0.019
Smoking history	4.286 (1.497,12.272)	0.011	0.889 (0.212,3.722)	0.872
NSE (ng/mL)	2.063 (1.141,3.731)	0.017	1.599(0.868,2.945)	0.132
MO(×109/L)	1.707 (1.023,2.851)	0.041	0.967(0.508,1.839)	0.819
PCCT spectral parameters
A-IC	0.655 (0.399,1.706)	0.097	-	-
A-NIC	0.466 (0.157,1.380)	0.163	-	-
A-40 keV	0.621 (0.375,1.029)	0.064	-	-
A-70 keV	0.519 (0.304,0.887)	0.016	-	-
A-100 keV	0.472 (0.269,0.827)	0.009	0.543(0.300,0.983)	0.044
A-λHU	0.690 (0.423,1.124)	0.136	-	-
A-Zeff	1.553 (0.923,2.611)	0.097	-	-
A-DEI	0.514 (0.271,0.975)	0.042	-	-
AEF	1.169 (0.719,1.899)	0.529	-	-
V-IC	0.519 (0.302,0.894)	0.018	-	-
V-NIC	0.713 (0.437,1.162)	0.174	-	-
V-40 keV	0.558 (0.331,0.941)	0.029	-	-
V-70 keV	0.546 (0.325,0.917)	0.022	-	-
V-100 keV	0.577 (0.347,0.958)	0.033	-	-
V-λHU	0.572 (0.340,0.963)	0.036	0.731(0.413,1.292)	0.281
V-Zeff	0.681(0.416,1.114)	0.152	-	-
V-DEI	1.152(0.678,1.958)	0.600	-	-
ECV	0.578(0.344,0.969)	0.038	-	-

A=arterial phase; AEF= arterial enhancement fraction; DEI=dual-energy index; ECV=extracellular volume; IC=iodine concentration; MO= monocyte count; NSE= neuron-specific enolase; NIC= normalized iodine concentration; V=portal venous phase; Zeff= effective atomic number; λHU=slope of the spectral HU curve

Note: P < 0.05. P-values were calculated from univariate and multivariate binary logistic regression analyses. Categorical variables (gender, smoking history) were analyzed with original dummy coding. All continuous variables were Z-standardized prior to regression; OR values of continuous variables represent the odds ratio per one standard deviation increment

Univariate binary logistic regression analysis was performed to investigate the associations between PCCT spectral parameters and 19Del as well as L858R mutations. No parameter reached statistical significance; however, the P values of A-40 keV, A-Zeff, and A-DEI were close to 0.05, with detailed results as follows: A-40 keV (P = 0.098, OR = 0.979, 95%CI: 0.954–1.004), A-Zeff (P = 0.088, OR = 1.234, 95%CI: 0.969–1.573), A-DEI (P = 0.084, OR = 0.001, 95%CI: 0.000–2.434). The results are summarized in Table S5.

Predictive performance of quantitative PCCT parameters for EGFR mutation

Arterial phase parameters A-70 keV, A-100 keV and A-DEI exhibited predictive value for EGFR mutation status, with area under the curve (AUC) values of 0.678 (P = 0.0058, 95% CI: 0.556–0.784), 0.693 (P = 0.0026, 95% CI: 0.572–0.797) and 0.642 (P = 0.0317, 95% CI: 0.519–0.752), respectively. The optimal cutoff values with corresponding sensitivity and specificity were as follows: A-100 keV > 48.55 (sensitivity: 55.56%, specificity: 85.71%) and A-DEI > 0.009 (sensitivity: 50.00%, specificity: 74.29%).Venous phase parameters V-IC, V-40 keV, V-70 keV, V-100 keV, and V-λHU showed predictive value for EGFR mutation status, with area under the curve (AUC) values of 0.688 (P = 0.0037, 95% CI: 0.567–0.792), 0.669 (P = 0.0094, 95% CI: 0.547–0.776), 0.649 (P = 0.0234, 95% CI: 0.527–0.759), 0.652 (P = 0.0207, 95% CI: 0.530–0.762), and 0.662 (P = 0.0137, 95% CI: 0.539–0.770), respectively. The optimal cutoff values with corresponding sensitivity and specificity were as follows: V-IC > 1.13 (sensitivity: 72.22%, specificity: 65.71%), V-40 keV > 147.5 (sensitivity: 61.11%, specificity: 71.43%), V-70 keV > 73.56 (sensitivity: 58.33%, specificity: 71.43%), V-100 keV > 51.96 (sensitivity: 66.67%, specificity: 62.86%), and V-λHU > 1.59 (sensitivity: 58.33%, specificity: 74.29%). In addition, ECV also exhibited predictive value for EGFR mutation status, with an AUC of 0.634 (P = 0.0468, 95% CI: 0.511–0.745). All other spectral parameters showed no predictive value (all P > 0.05). These results are presented in Table 4; Fig. 3.

Table 4.

ROC curve analysis of PCCT spectral parameters for predicting EGFR mutation in lung adenocarcinoma

Parameters	Raw AUC	Threshold	P	95% DeLong CI	Sensitivity(%)	Specificity(%)	Optimism-corrected AUC	95% BCa Bootstrap CI
A-IC	0.614	> 0.73	0.0900	0.491–0.727	47.22	74.29	0.606	0.473–0.740
A-NIC	0.605	> 0.066	0.1237	0.482–0.719	80.56	42.86	0.598	0.461–0.737
A-40 keV	0.630	> 97.82	0.0529	0.507–0.742	52.78	74.29	0.623	0.482–0.755
A-70 keV	0.678	> 55.12	0.0058	0.556–0.784	63.89	71.43	0.674	0.531–0.792
A-100 keV	0.693	> 48.55	0.0026	0.572–0.797	55.56	85.71	0.688	0.552–0.805
A-λHU	0.602	> 0.58	0.1346	0.479–0.717	83.33	42.86	0.589	0.461–0.735
A-Zeff	0.513	≤ 0.889	0.8524	0.391–0.634	25.00	94.29	0.527	0.367–0.654
A-DEI	0.642	> 0.009	0.0317	0.519–0.752	50.00	74.29	0.638	0.499–0.768
AEF	0.508	> 0.521	0.9094	0.387–0.629	55.56	54.29	0.502	0.362–0.641
V- IC	0.688	> 1.13	0.0037	0.567–0.792	72.22	65.71	0.680	0.484–0.754
V- NIC	0.625	> 0.4	0.0655	0.502–0.737	44.44	85.71	0.618	0.480–0.749
V- 40 keV	0.669	> 147.5	0.0094	0.547–0.776	61.11	71.43	0.661	0.526–0.787
V- 70 keV	0.649	> 73.56	0.0234	0.527–0.759	58.33	71.43	0.643	0.508–0.768
V- 100 keV	0.652	> 51.96	0.0207	0.530–0.762	66.67	62.86	0.645	0.508–0.774
V- λHU	0.662	> 1.59	0.0137	0.539–0.770	58.33	74.29	0.658	0.520–0.781
V-Zeff	0.626	> 8.1	0.0591	0.503–0.738	55.56	68.57	0.617	0.480–0.752
V-DEI	0.631	> 0.012	0.0526	0.508–0.742	55.56	71.43	0.626	0.484–0.754
ECV	0.634	> 0.173	0.0468	0.511–0.745	69.44	60.00	0.625	0.494–0.757
Clinical factors	0.767	≤ 0.5559	< 0.0001	0.642–0.865	70.00	75.00	0.762	0.612–0.870
PCCT spectral parameters	0.713	≤ 0.48277	0.0006	0.593–0.814	66.67	74.29	0.707	0.580–0.825
Combined model	0.771	≤ 0.54442	< 0.0001	0.656–0.863	80.56	74.29	0.768	0.643–0.871

A=arterial phase; AEF= arterial enhancement fraction; ECV=extracellular volume; DEI=dual-energy index; IC=iodine concentration; NIC= normalized iodine concentration; V=portal venous phase; Zeff= effective atomic number; λHU= slope of the spectral HU curve; ROC=receiver operating characteristic; AUC= area under the curve; 95% CI=95% DeLong confidence interval

Note: P < 0.05, the P-value is derived from the receiver operating characteristic (ROC) curve analysis

Fig. 3.

ROC curves of arterial phase PCCT spectral parameters (A), venous phase PCCT spectral parameters (B), clinical parameters (C), and three predictive models (D) for predicting EGFR gene mutation in patients with lung adenocarcinoma

The predictive ability of clinical characteristics, PCCT spectral parameters, and their combination for EGFR mutation status was evaluated using ROC curve analysis with bootstrap internal validation (1 000 resampling iterations). The clinical model, which included sex, smoking history, NSE and MO, achieved an AUC of 0.767 (95% CI: 0.642–0.865, P < 0.0001), with an optimism-corrected AUC of 0.762 (95% BCa bootstrap CI: 0.612–0.870). The PCCT model, incorporating A-100 keV and V-λHU, had an AUC of 0.713 (95% CI: 0.593–0.814, P = 0.0006), with an optimism-corrected AUC of 0.707 (95% BCa bootstrap CI: 0.580–0.825). The combined model integrating clinical characteristics and PCCT spectral parameters achieved an original AUC of 0.771 (95% CI: 0.656–0.863, P < 0.0001), with an optimism-corrected AUC of 0.768 (95% BCa bootstrap CI: 0.643–0.871). DeLong’s test revealed no statistically significant difference in predictive performance between the combined model, the PCCT model and the clinical model, indicating that the combined model was not superior to the clinical model and the PCCT model (all P > 0.05). The results are shown in Fig. 3.

ROC curve analysis revealed that A-40 keV and A-DEI enabled the differentiation between 19Del and L858R mutations. For A-40 keV, the original AUC was 0.694 (95% CI: 0.507–0.844, P = 0.0472), with an optimism-corrected AUC of 0.688. For A-DEI, the original AUC was 0.704 (95% CI: 0.517–0.851, P = 0.0340), with an optimism-corrected AUC of 0.697. All other PCCT spectral parameters showed no statistically significant differences (all P > 0.05). The results are summarized in Table S6 and Fig. S1.

Discussion

Lung adenocarcinoma exhibits high genetic heterogeneity, and EGFR mutation is a core biomarker for targeted therapy. As a novel high-resolution spectral imaging modality, PCCT can more accurately and noninvasively evaluate EGFR-related tumor microvascular perfusion features than conventional CT and spectral CT [15, 16]. This study analyzed PCCT parameter differences between EGFR-mutant and wild-type lung adenocarcinoma, as well as between 19Del and L858R subtypes, to explore its value in noninvasive genetic prediction and support individualized targeted therapy.

In our study, the EGFR mutation rate among patients with lung adenocarcinoma was 51.4%, which was consistent with the approximately 50% mutation rate reported in Asian populations [17]. Serum NSE levels were significantly lower in EGFR-mutant patients than wild-type counterparts, and univariate logistic regression revealed that higher serum NSE was correlated with EGFR wild-type status. This observation was inconsistent with the results from Jiang et al. [18], who reported elevated NSE levels in EGFR-mutant individuals. Apart from cohort differences in pathological composition and TNM stage distribution, the wild-type group in our study contained a substantially higher proportion of males and smokers. The resulting confounding baseline effect, where heavy smoking increases baseline NSE concentrations [19], largely accounts for the divergent findings between the two studies. Furthermore, EGFR mutations were more frequently observed in female and never-smoking patients, in line with previous findings [20, 21]. Among the above clinical characteristics, multivariate binary logistic regression analysis identified sex as an independent predictive factor for EGFR mutation status. Female patients were more prone to EGFR mutations, which is closely related to the fact that most female patients are non-smokers.

The results of the present study demonstrated that arterial phase A‑70 keV, A‑100 keV, A‑DEI, as well as venous phase V‑40 keV, V‑70 keV, V‑100 keV, V‑λHU and V‑DEI were significantly higher in patients with EGFR mutation‑positive lung adenocarcinoma than in those with wild‑type EGFR (P < 0.05), which was generally consistent with previous studies [10, 22]. The underlying mechanism may be that EGFR‑mutant tumors exhibit higher microvascular density and more abundant blood perfusion, which is highly consistent with the biological characteristic that the EGFR signaling pathway promotes tumor angiogenesis by upregulating pro‑angiogenic factors such as VEGF [23, 24]. Univariate regression analysis further demonstrated that these PCCT parameters were significantly associated with EGFR mutation status, suggesting their potential value as non-invasive imaging indices for evaluating EGFR mutation status in lung adenocarcinoma. Multivariate regression analysis showed that arterial phase 100 keV (A‑100 keV) was an independent predictive factor for EGFR mutation. A possible reason is that, compared with low keV levels (40–70 keV), which are more sensitive to tissue density, calcification and necrosis and tend to be confounded by non‑vascular components, 100 keV is close to the effective energy of conventional 120 kVp CT, with stronger X‑ray penetration and fewer beam‑hardening artifacts. The CT value at this energy level mainly depends on tissue iodine concentration, allowing a more reliable reflection of actual tumor microvascular perfusion and iodine uptake.

ROC curve analysis showed that arterial phase parameters (A-70 keV, A-100 keV and A-DEI) and venous phase parameters (V-IC, V-40 keV, V-70 keV, V-100 keV, V-λHU) and ECV had favorable predictive performance for EGFR mutation, which was generally consistent with part of the findings reported using conventional spectral CT to predict EGFR mutation in lung adenocarcinoma [11, 12]. Predictive model performance is modulated by numerous confounding variables; cross-platform comparisons of diagnostic metrics therefore carry limited clinical interpretive value. Nonetheless, recent phantom studies have provided objective evidence that PCCT achieves superior spectral quantification relative to conventional energy-integrating detector CT (EID-CT), delivering images with lower noise, finer texture, and material quantification with reduced bias and greater reproducibility. Winfree et al. [25] demonstrated that after modulation transfer function spatial resolution matching, single-source PCCT yielded a 22% reduction in iodine map noise relative to dual-source EID-CT under identical radiation dose, while dual-source PCCT achieved a 39–41% noise reduction. Greffier et al. [26] demonstrated that PCCT could achieve a 68–91% radiation dose reduction relative to standard-dose dual-source energy-integrating CT (11 mGy) while retaining equivalent lesion detectability on low-keV virtual monoenergetic images. These phantom data confirm that PCCT’s improved spectral discrimination and reduced electronic noise may yield more robust spectral quantification in clinical oncologic imaging and theoretically confer advantages in model stability and generalizability compared with conventional spectral CT.

In the present study, further analysis was conducted on the predictive capability of PCCT parameters, clinical characteristics, and their combination, demonstrating that all three models could predict EGFR status. The combined model yielded the numerically highest AUC; nevertheless, DeLong’s test revealed no statistically significant differences across the three models (P > 0.05). These results indicate that PCCT parameters may serve as auxiliary indicators for identifying patients with EGFR-mutated lung adenocarcinoma and may provide supplementary evidence to support individualized targeted-therapy decision-making. From a clinical-utility perspective, the predictive performance of the established models in our cohort (AUC values around 0.7, with most sensitivity and specificity ranging from 55% to 80%) indicates that these models are not adequate to independently guide clinical decision-making for EGFR-targeted therapy. Therefore, PCCT parameters should be regarded only as a complementary and adjunctive tools and are not a substitute for gold-standard tissue- or blood-based EGFR testing, which remains the definitive reference for identifying EGFR mutation status.

In the subgroup analysis of EGFR mutation subtypes, PCCT spectral parameters were generally higher in the 19Del group than in the L858R group, with only A-DEI showing a statistically significant difference (P = 0.049). ROC curve analysis further demonstrated that both A-40 keV (P = 0.0472) and A-DEI (P = 0.0340) possessed statistically significant discriminatory performance for subtype classification, with original AUC values of 0.694 and 0.704, and optimism-corrected AUC values of 0.688 and 0.697, respectively. We hypothesize that these findings stem from biological heterogeneity between the two subtypes. Despite reports of higher microvascular density and greater angiogenic activation in L858R tumors [23, 27], the 19Del group paradoxically showed higher A-DEI. One explanation is that arterial iodine uptake may depend more on vascular functional integrity than on vessel quantity. L858R-driven neovascularization may be immature and hyperpermeable [27], causing rapid contrast extravasation and lower A-DEI. In contrast, microvessels in 19Del tumors may be more mature and functionally stable, enabling better contrast retention and higher A-DEI. A-40 keV is sensitive to changes in iodine concentration, while A-DEI integrates multi-energy level attenuation differences. Both parameters can capture the perfusion heterogeneity between the two subtypes and thus present favorable discriminatory efficacy. The above mechanistic speculations still need to be verified by further experiments. Of note, these subtype-related findings are preliminary due to the limited sample size and require validation in larger cohorts.

This study has several limitations. First, as a single-center retrospective study with a limited sample size, selection bias may exist, and the generalizability of our results requires further validation in multicenter, large-sample prospective studies. Second, several tumor characteristics — including tumor size, location, histologic subtype, degree of differentiation, pleural invasion, nodal status, and the proportion of solid versus ground-glass components — were not collected, and therefore could not be adjusted for as potential confounders in the multivariate analysis. Although ROI placement was restricted to solid tumor components with necrosis, cystic change, and calcification manually excluded, the inherent fraction of ground-glass component within the overall tumor may still influence iodine-based spectral quantification. The residual confounding bias arising from these unmeasured variables represents an important limitation of this study. Third, we did not explore the detailed molecular mechanisms underlying the associations between PCCT spectral parameters and EGFR mutation-related signaling pathways (e.g., the VEGF pathway), and the related mechanistic hypotheses still need to be verified by basic experiments. Finally, due to the limited sample size, only the two common EGFR mutation subtypes (19Del and L858R) were analyzed, while other rare subtypes were not included. Thus, our conclusions still need to be validated with an expanded sample size.

Conclusion

In conclusion, quantitative PCCT spectral parameters non-invasively predict EGFR mutation in lung adenocarcinoma, with female sex and A-100 keV as independent predictors. Though the clinical-PCCT combination yielded a marginally higher AUC, all models showed modest predictive performance and DeLong’s test revealed no inter-model statistical differences. Therefore, PCCT parameters should be regarded as auxiliary imaging markers rather than standalone diagnostic criteria. A-40 keV and A-DEI tentatively distinguish 19Del from L858R subtypes, yet these exploratory subgroup results require validation in larger cohorts. PCCT provides supplementary imaging evidence to guide individualized lung adenocarcinoma treatment decisions.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

JR and YJ came up with the design and conception. The PCCT imaging analysis was performed by LL and JR. The data statistic was conducted by JR and LL. The original writing of the draft and its editing were by YH and YY. All authors contributed to the study, reviewed the data analysis, and approved the final version of the manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (Grant Nos. 12275162 and 12575365).

Data availability

The datasets used and analyzed in this study are available from the corresponding author upon request.

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. This study was approved by the Ethics Committee of the Cancer Hospital Affiliated to Shandong First Medical University. Written informed consent was waived due to the retrospective design of this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.