Diagnostic Performance of MRI for the Detection of Pulmonary Nodules: A Systematic Review and Meta-Analysis

Abstract

Purpose

To perform a meta-analysis of the diagnostic performance of MRI for the detection of pulmonary nodules, with use of CT as the reference standard.

Materials and Methods

PubMed, Embase, Scopus, and other databases were systematically searched for studies published from January 2000 to March 2023 evaluating the performance of MRI for diagnosis of lung nodules measuring 4 mm or larger, with CT as reference. Studies including micronodules, nodules without size stratification, or those from which data for contingency tables could not be extracted were excluded. Primary outcomes were the per-lesion sensitivity of MRI and the rate of false-positive nodules per patient (FPP). Subgroup analysis by size and meta-regression with other covariates were performed. The study protocol was registered in the International Prospective Register of Systematic Reviews, or PROSPERO (no. CRD42023437509).

Results

Ten studies met inclusion criteria (1354 patients and 2062 CT-detected nodules). Overall, per-lesion sensitivity of MRI for nodules measuring 4 mm or larger was 87.7% (95% CI: 81.1, 92.2), while the FPP rate was 12.4% (95% CI: 7.0, 21.1). Subgroup analyses demonstrated that MRI sensitivity was 98.5% (95% CI: 90.4, 99.8) for nodules measuring at least 8–10 mm and 80.5% (95% CI: 71.5, 87.1) for nodules less than 8 mm.

Conclusion

MRI demonstrated a good overall performance for detection of pulmonary nodules measuring 4 mm or larger and almost equal performance to CT for nodules measuring at least 8–10 mm, with a low rate of FPP.

Systematic review registry no. CRD42023437509

Keywords: Lung Nodule, Lung Cancer, Lung Cancer Screening, MRI, CT

Supplemental material is available for this article.

© RSNA, 2024

Summary

MRI had good sensitivity for the detection of pulmonary nodules measuring 4 mm or larger and almost equivalent sensitivity to CT for the detection of nodules measuring 8–10 mm.

Key Points

• ■ In a meta-analysis of 10 studies including 1354 patients and 2062 CT-detected pulmonary nodules, MRI had a sensitivity of 87.7% for the detection of nodules measuring 4 mm or larger.

• ■ The sensitivity of MRI for nodules measuring at least 8–10 mm was 98.5%.

• ■ Patients with negative MRI findings were reliably classified as free of pulmonary nodules compared with CT, with a rate of false-positive nodules per patient of 12.4%.

Introduction

Malignant nodules may arise from either primary lung cancer or distant metastasis from other primary sites. Primary lung cancer is the second most common malignancy and the leading cause of cancer-related death worldwide (1). The lung is also one of the most common sites for distant metastasis from other primary neoplasia. Therefore, the detection of lung nodules at imaging is essential for lung cancer screening and metastatic staging of other malignancies.

CT is currently the reference standard modality for detection of pulmonary nodules given its availability, excellent contrast resolution in the lung, and speed of image acquisition (2); however, CT has two major limitations. First is the use of ionizing radiation, especially in oncologic patients who often already have a high radiation burden from prior testing or follow-up. Extrapolating from data derived from nuclear accidents, one study estimates that CT may be responsible for up to 2% of all future cancer cases in the United States, given the rising rates of imaging over the past 3 decades (3). Second is that CT has limited capacity to predict the likelihood of malignancy of a subset of nodules, given that benign nodules are particularly common in populations undergoing lung cancer screening or cancer staging (4). In this context, MRI serves as an alternative to CT for detection of pulmonary nodules given the lack of ionizing radiation and its ability to differentiate malignant and benign nodules with similar or better accuracy compared with PET/CT (5,6). Moreover, a recent study showed that MRI has a favorable cost-effectiveness compared with low-dose CT (LDCT) driven by fewer false-positive screening examinations (7).

Although MRI overcomes some limitations of CT, the sensitivity of the method for detection of lung nodules is unclear. The aim of this systematic review and meta-analysis was to evaluate the diagnostic performance of MRI for the detection of lung nodules measuring 4 mm or larger with use of CT as the reference standard.

Materials and Methods

This systematic review followed the Preferred Reporting Items for Systematic Reviews of Diagnostic Test Accuracy Studies, or PRISMA-DTA, guidelines (8) and was registered in the International Prospective Register of Systematic Reviews database, or PROSPERO (registration no. CRD42023437509).

Study Selection

The available literature published from January 2000 to March 2023 was searched in the MEDLINE, Embase, Scopus, BVS, and Cochrane databases using the comprehensive terms included in Appendix S1. Additional references were searched by cross-checking bibliographies of relevant eligible studies.

We included studies that evaluated the performance of MRI for the detection of pulmonary nodules measuring 4 mm or larger in the context of lung cancer screening, incidental nodule follow-up, or cancer diagnosis or staging, with CT as the reference standard. The following exclusion criteria were used: (a) studies from which contingency tables could not be extracted; (b) studies including micronodules (≤3 mm) without stratified analysis by size to allow extraction of a separate contingency tables for nodules measuring 4 mm or larger; (c) studies performed in patients younger than 18 years; (d) studies that evaluated integrated PET with MRI; (e) studies focused on nodules in infectious or inflammatory conditions; (f) studies with an interval between CT and MRI of more than 30 days; (g) studies with less than 10 nodules or patients; (h) studies not published in English; and (i) conference abstracts, case series, or preprints.

Data Extraction

Two authors (C.C.C. and S.A.) reviewed all selected abstracts and titles for inclusion criteria, with any discrepancies resolved by a third author (B.H.). The final selection of full texts was then reviewed in the same manner to confirm eligibility. Contingency tables for both per-lesion and per-patient diagnostic performance were collected. In studies with more than one sequence being investigated, the sequence with the highest diagnostic performance was used for the contingency tables in case an overall reference performance using all available sequences was not provided by the authors. Additional demographics and characteristics of each study, including year of publication, study protocol, and nodule size, among others, were extracted accordingly.

Study Quality Assessment

Two reviewers (C.C.C. and S.A.) independently assessed the quality of eligible studies, with any discrepancies resolved by a third reviewer (B.H.). The Quality Assessment of Diagnostic Accuracy Studies, or QUADAS-2, checklist was used to assess the methodologic quality of the included studies (9). This quality control instrument consists of four domains: patient selection, index testing, reference standard, and flow and timing. Each was assessed in terms of risk of bias and the first three in terms of concerns regarding applicability.

Data Synthesis and Statistical Analysis

The primary end points of our study were the per-lesion sensitivity of MRI compared with the reference standard, CT, for the detection of lung nodules measuring 4 mm or larger and the rate of false-positive nodules per patient (FPP). Subgroup analyses based on nodule size according to a cutoff of 8–10 mm was also performed. The pooled sensitivities and FPP rate with their respective 95% CIs were calculated using a random-effect analysis.

The heterogeneity of the articles was further assessed by visually inspecting forest plots and performing χ² tests (with P < .10 indicating heterogeneity). We used the I² statistic to quantify inconsistencies among the studies, with an I² > 50% indicating substantial heterogeneity. Multivariate meta-regression was performed for subgroup analyses based on three covariates: indication of imaging (lung cancer screening vs others), magnetic field strength (1.5 vs 3.0 T), and MRI technique (section thickness ≤1.25 vs >1.25 mm). The significance level was set at .05 for all analyses. The software R (version 4.2.3, R Core Team 2023) was used to conduct this meta-analysis.

Results

Study Selection and Characteristics

The initial search returned 638 studies, of which 135 were duplicates (Fig 1). Additionally, 14 studies were found using references from other included articles. Among the 517 studies that underwent title and abstract screening, a total of 45 were retrieved for full-text review and screened for eligibility, of which 10 studies met the inclusion criteria (10–19). In total, the meta-analysis comprised 1354 patients and 2062 nodules detected at CT. Most domains of QUADAS-2 were found to be at low risk of bias for the included studies (Fig S1).

Figure 1:

Preferred Reporting Items for Systematic Reviews and Meta-Analyses, or PRISMA, flow diagram.

The characteristics of the included studies are summarized in Table 1. All studies were prospective and single-center, except for three multicenter studies (14,18,19). The interval between CT and MRI was less than 14 days for all studies. Regarding the indication for imaging, three studies were performed in the context of lung cancer screening programs (13,18,19) (the one by Li and Zhu et al [19] was performed in a chronic obstructive pulmonary disease screening context), while three studies were performed for lung cancer diagnosis and three for metastatic staging of other malignancies. The indication for imaging is unclear in the study by Cieszanowski et al (10). There was a significant variability in the cutoffs used by studies for the “higher risk” nodules, with some studies using 8 mm while others used 10 mm to stratify the groups. Therefore, we stratified the risk categories using a cutoff above or below 8–10 mm for binary classification of all studies. Among the eight studies that performed stratified diagnostic performance analysis by size, the median proportion of nodules above 8–10 mm was 28% (range, 12%–60%). Only two studies used LDCT as the reference standard for MRI comparison. For most studies, the diagnostic performance of MRI was obtained from a consensus of two radiologists, except for the study by Dewes et al (11), in which the performance was derived from an average sensitivity of three radiologists.

Table 1:

Characteristics of the 10 Included Studies

Only three of the studies clearly reported the type of nodule included and the prevalence of ground-glass nodules and part-solid nodules (Table 1), with the two studies by Ohno et al (14,18) reporting 24% and 17%, respectively, of total nodules as ground-glass nodules and 12% in both studies as part-solid nodules. However, most of the remaining authors did not report excluding ground-glass nodules and part-solid nodules. Li and Zhu et al (19) reported 2% of total nodules as part-solid nodules, and the ground-glass nodules were excluded from this meta-analysis due to lack of details about size.

The details on magnet strength, best-performing MRI sequence, and section thickness of each individual study are summarized in Table 2. All included studies used either a 1.5- or 3.0-T MRI scanner. Most of the studies performed more than one MRI sequence, with the best-performing sequence and section thickness provided in Table 2. Only three studies used a thin section thickness (<1.25 mm). These studies also used a radial acquisition, while most other studies used a cartesian acquisition. Only the two studies by Ohno et al (14,18) used an ultrashort echo time sequence.

Table 2:

Details on MRI Scan Acquisition Parameters for Each Study

Diagnostic Performance and Subgroup Analysis

The pooled per-lesion sensitivity of MRI to detect lung nodules measuring 4 mm or larger was 87.7% (95% CI: 81.1, 92.2), with significant heterogeneity in the analysis (I² = 93%) (Fig 2). All 10 studies reported the overall false-positive rate of MRI for false-positive nodules of any size, with a pooled FPP rate of 12.4% (95% CI: 7.0, 21.1) (Fig 3), which would approximate to an estimated overall specificity of 88%.

Figure 2:

Forest plot shows the pooled sensitivity of MRI for detection of pulmonary nodules measuring 4 mm or larger. FN = false negative, TP = true positive.

Figure 3:

Forest plot shows the pooled rate of false-positive (FP) nodules per patient for pulmonary nodules of any size.

At visual analysis, it was observed that the study by Heye et al (12) presented the lowest sensitivity and contributed the most to heterogeneity in the overall group. For the FPP analysis, the study by Ohno et al (14) presented the highest FPP rate, contributing the most to the heterogeneity in that analysis. In the analysis stratified by nodule size, the sensitivity of MRI for nodules below 8–10 mm was 80.5% (95% CI: 71.5, 87.1) (Fig 4), while heterogeneity remained elevated (I² = 82%). When considering only nodules larger than 8–10 mm, the sensitivity of MRI increased to 98.5% (95% CI: 90.4, 99.8), with no heterogeneity (Fig 5).

Figure 4:

Forest plot shows the pooled sensitivity of MRI for detection of pulmonary nodules under 8–10 mm. FN = false negative, TP = true positive.

Figure 5:

Forest plot shows the pooled sensitivity of MRI for detection of pulmonary nodules larger than 8–10 mm. FN = false negative, TP = true positive.

Subgroup analyses to explore additional factors contributing to the heterogeneity are demonstrated in Table 3. MRI technique (≤1.25-mm section thickness) and studies performed outside the context of lung cancer screening showed higher overall sensitivity (P = .04 and .01, respectively). MRI magnetic field strength was not associated with a significant difference in sensitivity (P = .16). None of the analyzed factors showed a significant impact on the rate of FPP.

Table 3:

Meta-Regression and Subgroup Analyses of MRI Diagnostic Performance for Nodules Measuring 4 mm or Larger

Discussion

This study demonstrated that the sensitivity of MRI was 87.7% for lung nodules measuring 4 mm or larger, which is widely accepted as the minimal size cutoff to be relevant for most patients undergoing lung cancer screening, given only new nodules above this cutoff would qualify for a recall of earlier than 12 months (20,21). For nodules measuring at least 8–10 mm, usually considered of higher risk and warranting closer follow-up or histologic analysis (21), the sensitivity of MRI was virtually equivalent to that of standard-dose CT, although only two studies compared MRI performance with that of LDCT. When comparing the sensitivity of MRI and LDCT for the detection of lung nodules in lung cancer screening, Ohno et al (18) found that both modalities had equivalent sensitivities (87.9% vs 87.1%, respectively) compared with standard-dose CT, which is supported by the results of our study.

The same study by Ohno et al (18) revealed that although both methods are equivalent, MRI had a slightly lower rate of FPP compared with LDCT (18% vs 27%; test for significance not performed). We encountered similar overall results in this meta-analysis with an overall FPP rate of 12.4%. The FPP rate can be seen as a proxy for per-patient specificity, which could be estimated as 88% (specificity = 1 – FPP rate). The reason for choosing the rate of FPP as a proxy for specificity lies in the difficulty of measuring a “true-negative” nodule (ie, the same nodule being negative at CT and MRI), which is possible in only a per-lobe or per-patient analysis. However, only three studies included healthy controls, given that most studies were performed in the context of lung cancer screening and cancer staging, in which patients rarely do not have any nodule to be assessed. According to the findings of our study, one would expect 12 false-positive nodules for every 100 patients undergoing MRI. In this case, the uncertainty lies in how many patients would be affected, as one could have more than one of the 12 false-positive nodules, in which case per-patient specificity is being underestimated.

Among the 1258 nodules that could be stratified by size, 882 (70%) were in the category of size less than 8–10 mm, with a sensitivity of around 80%. This implies that about 172 of these smaller nodules went undetected at MRI. However, studies that analyzed stricter size stratification within this category (eg, 4–6 mm vs 6–8 mm) achieved sensitivities higher than 90% in the latter subgroup. This was particularly evident in studies using MRI techniques with thinner section thickness, such as those by Ohno et al (14,18) and Yu and Yu et al (17).

Although MRI is unlikely to replace LDCT for lung cancer screening on a large scale, MRI offers two inherent advantages over CT. First is the lack of ionizing radiation: According to the American College of Radiology, the number of procedures using ionizing radiation that a patient can undergo throughout life and the amount of radiation to which each of these procedures exposes patients has largely increased (22). Therefore, MRI could be seen as an alternative particularly for patients with a high radiation burden. Second is the potential for malignancy stratification beyond growth, nodule type, and characteristics such as spiculation, particularly when coupled with diffusion-weighted imaging (DWI). Although official guidelines do not yet provide specific recommendations in this regard (23), there is compelling evidence suggesting that MRI with DWI could be used as an alternative to PET/CT for malignancy stratification in indeterminate lung nodules. A recent meta-analysis (5) has shown that DWI had 83% sensitivity and 91% specificity for the differentiation of malignant and benign nodules, which can add more value in lung cancer screening for patients with nodules classified as Lung Imaging Reporting and Data System, or Lung-RADS, categories 3 and 4A, instead of patients undergoing an additional short follow-up LDCT examination (5).

One of the potential concerns of MRI in lung cancer screening lies in the detection of ground-glass and part-solid nodules. Among the studies in our review, only three consistently reported the prevalence and diagnostic accuracy in these nodules, with sensitivity being very similar (14,18) to considerably lower (19) than that of solid nodules. These differences are likely attributed to differences in technique, with the first two studies (14,18) using a radial ultrashort echo time sequence with a 1-mm section thickness and the other (19) using a cartesian volumetric interpolated breath-hold examination, or VIBE, acquisition with 4-mm section thickness. Additionally, the agreement of CT and MRI for the classification of a nodule into solid, ground-glass, or part-solid was addressed only by Ohno et al (18), showing a slight tendency to classify some ground-glass opacities as ground-glass nodules and some ground-glass nodules as solid. This was attributed to increased proton density within ground-glass nodules due to volume loss of small subsolid nodules at end-tidal volume at MRI (18). Nevertheless, given that this was mostly observed for nodules smaller than 6 mm in that study, this difference did not significantly affect the Lung-RADS classification.

Another potential concern is the morphologic assessment of solid nodules at MRI, such as spiculation and cavitation, as these features have been shown to determine the likelihood of malignancy based on CT (24). The group using the radial VIBE acquisition with thin sections found a 100% agreement between CT and MRI for cavitation, a moderate agreement for spiculation (κ = 0.60), and a lower agreement for calcifications (κ = 0.37) (25).

MRI also has higher equipment acquisition and maintenance costs when compared with CT. In this context, recent advances in low-field-strength closed-bore MRI technology have the potential to address this issue while also offering the benefit of lower susceptibility artifacts at air-tissue interfaces (26). Although this was not an exclusion criterion in our study, we did not find any article using lower field strength that met inclusion criteria during our systematic searches. However, a more recent study using less expensive modern low-field-strength MRI (0.55 T) demonstrated excellent sensitivity in the detection of pulmonary nodules when compared with CT, achieving a detection accuracy of 100% for nodules measuring 6 mm or larger and 80% for those measuring 4 mm or smaller and smaller than 6 mm (27).

MRI could be used not only for lung cancer screening, but also for the metastatic staging of other primary cancers as a one-stop study for patients who would undergo an abdominal MRI examination for other indications. The main limitations at this stage for applicability of MRI to evaluate distant pulmonary metastasis is that missing nodules less than 4 mm would most likely impact patient management, as opposed to lung cancer screening, in which nodules smaller than 4 mm are considered benign (20). However, the advantage of MRI compared with CT in this scenario relies on the stratification of malignancy risk of pulmonary nodules, especially when combined with DWI as previously demonstrated in the literature (6). Our study is also important to support further investigation of lung MRI for cancer staging, given that some primary cancers, such as pancreatic and colorectal cancers, have also been shown to benefit from the addition of liver MRI over CT alone for hepatic staging (28,29).

Our study had many limitations, including those inherent to the meta-analysis of diagnostic studies, namely heterogeneity. However, after controlling for nodule size and other parameters in our meta-regression, we were able to identify potential factors to explain the heterogeneity. The study samples were also diverse, with some studies including lung cancer screening programs while the others were conducted in the setting of lung cancer diagnosis or staging of other primary cancer. MRI techniques have evolved over the years and were variable in the included studies, which affects interpretation. As shown in our sensitivity analysis, studies using sequences with T1 weighting, thin sections, and noncartesian acquisitions had superior performance. Last, given that included studies evaluated per-lesion diagnostic accuracy, per-patient specificity was difficult to obtain; although the FPP rate was used as a proxy, it likely underestimates the true specificity.

In summary, we found high sensitivity of MRI for lung nodules measuring 4 mm or larger when using CT as the reference standard. The sensitivity of MRI for nodules measuring at least 8–10 mm was comparable to that of CT, with MRI having the advantages of being a nonionizing radiation method and being able to provide risk stratification for nodules in this category. Although it is unlikely that MRI will replace CT as the main modality for lung nodule screening, our results should encourage further investigation of MRI as a radiation-free alternative in lung cancer screening and metastatic staging of other malignancies as a one-stop study.