Skip to main content
Download PDF

T1 mapping-based radiomics in the identification of histological types of lung cancer: a reproducibility and feasibility study

BMC Medical Imaging volume 24, Article number: 308 (2024) Cite this article

Abstract

Background

T1 mapping can quantify the longitudinal relaxation time of tissues. This study aimed to investigate the repeatability and reproducibility of T1 mapping radiomics features of lung cancer and the feasibility of T1 mapping-based radiomics model to predict its pathological types.

Methods

The chest T1 mapping images and clinical characteristics of 112 lung cancer patients (54 with adenocarcinoma and 58 with other types of lung cancer) were collected retrospectively. 54 patients underwent twice short-term T1 mapping scans. Regions of interest were manually delineated on T1 mapping pseudo-color images to measure the mean native T1 values of lung cancer, and radiomics features were extracted using the semi-automatic segmentation method by two independent observers. The patients were randomly divided into training group (77 cases) and validation group (35 cases) with the ratio of 7:3. Interclass correlation coefficients (ICCs), Student’s t-test or Mann-Whitney U tests and least absolute shrinkage and selection operator (LASSO) were used for feature selection. The optimum features were selected to establish a logistic regression (LR) radiomics model. Independent sample t-test, Mann Whitney U-test or chi square test were used to compare the differences of clinical characteristics and T1 values. Performance was compared by the area under the receiver operating characteristic (ROC) curve (AUC).

Results

In the training group, smoking history, lesion type and native T1 values were different between adenocarcinoma and non-adenocarcinoma patients (P = 0.004–0.038). There were 1035 (54.30%) radiomics features meet the intra-and inter-observer, and test-retest reproducibility with ICC > 0.80. After feature dimensionality reduction and model construction, the AUC of T1 mapping-based radiomics model for predicting the pathological types of lung cancer was 0.833 and 0.843, respectively, in the training and validation cohorts. The AUCs of T1 value and clinical model (including smoking history and lesion type) were 0.657 and 0.692 in the training group, and 0.722 and 0.686 in the validation group. Combined with T1 mapping radiomics, clinical model and T1 value to establish a combined model, the prediction efficiency was further improved to 0.895 and 0.915 in the training and validation groups.

Conclusions

About 50% of the T1 mapping-based radiomics features displayed relatively poor repeatability and reproducibility. While T1 mapping-based radiomics model is valuable in identification of histological types of lung cancer despite the measurement variability. Combination of T1 mapping radiomics model, clinical characteristics and native T1 value can improve the predictive value of pathological types of lung cancer.

Peer Review reports

Background

Lung cancer is one of the most common malignant tumours [1]. Among them, adenocarcinoma is the most common pathological type. It is important to distinguish adenocarcinoma from other types of lung cancer prior to initiating treatment because many patients with adenocarcinoma have gene mutations and targeted therapy is recommended, while patients with squamous cell carcinoma and small-cell lung cancer generally choose radiotherapy, chemotherapy or immunotherapy [2]. Pathology obtained by surgery, puncture and bronchoscopy is the gold standard of clinical diagnosis. However, these procedures are invasive, may result in false negatives, and cannot be repeatedly performed. In traditional imaging, some scholars [3] have suggested that lesion type, size, spiculation and pleural indentation signs as well as obstructive pneumonia and atelectasis may have the potential to distinguish the histological types of lung cancer. However, these features are based on visual assessment and manual measurements, which are subjective, unstable and overlap.

Radiomics is an alternative technique that employs high-throughput quantitative imaging features from traditional images for clinical diagnosis [4]. Computed tomography (CT) is the most commonly used in clinic. Nevertheless, CT has the risk of radiation exposure, especially for patients who need repeated examinations within a short time interval. Magnetic resonance imaging (MRI) is absence of radiation exposure, which also has shown a potential value in the diagnosis and prediction of histological grade, type and epidermal growth factor receptor mutation status of lung cancer [5,6,7,8,9]. Xie et al. [5] found that combining multi-sequence radiomics, clinical and imaging features, the predictive efficacy was significantly improved for discriminating between benign and malignant solid pulmonary nodules or masses. However, previous studies have mainly focused on T1-weighted imaging (T1WI), T2WI, diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) mapping. Conventional MR images do not directly measure the longitudinal relaxation time (T1) of the tissue, but indirectly reflect it as signal intensity (SI). T1 mapping quantifies T1 values of tissues, and has become an important imaging tool for characterization of myocardial tissue [10]. Recently, studies [11,12,13] have confirmed that the clinical application of T1 mapping in lung cancer patients is feasible, and the native T1 values of lung lesions have a potential value in the diagnosis and differentiation of pathological type of lung cancer. In addition, Yan et al. [14] found that T1 mapping based radiomics model can potentially distinguish benign and malignant lung lesions. However, the feasibility of T1 mapping-based radiomics in the identification of histological types of lung cancer is still not well estimated.

The imaging features are highly redundant, and lung MR images may be influenced by respiratory, heart pulsation and susceptibility artefacts. Yan et al. [14] reported that only 21/107 T1 mapping-based radiomics features show relatively good interobserver agreement with intra-class correlation coefficients (ICCs) greater than 0.6, and Peerlings et al. [15] reported that only 25% of ADC radiomics features present relatively satisfactory test-retest reproducibility with concordance correlation coefcients (CCC) greater than 0.85. In Jiang [13] et al. study, the mean native T1 value is a repeatable and reproducible parameter. So, it is necessary to exclude the unstable features when establishing a radiomics model. ICC greater than 0.80 is a commonly used repeatability criterion in radiomics research [16]. But the repeatability and reproducibility of T1 mapping radiomics features of lung cancer haven’t been studied yet.

So we aimed to investigate the repeatability and reproducibility of radiomics features extracted from T1 mapping for selecting stable features with ICC > 0.80, and to further explore the feasibility of T1 mapping-based radiomics models to identify the histological types of lung cancer. So as to establish an optimal predictive model by integrating clinical features, T1 values and T1 mapping radiomics for predicting the histological types of lung cancer.

The main contributions of this paper can be summarized as follows:

  1. 1.

    There were 1035 (54.30%) T1 mapping radiomics features were stable between repeated measurements and scans, suggesting that the reproducibility of radiomics features needs to be evaluated when establishing a radiomics model.

  2. 2.

    T1 mapping-based logistic regression radiomics model is feasible in the identification of histological types of lung cancer.

  3. 3.

    The proposed model integrating clinical features, T1 values and T1 mapping radiomics offers a rapid, non-invasive method for early characterizing lung cancer, which can be extended to the identification of other tumours.

The article was structured as follows: The Patients and methods section detailed the study participants, MRI protocols, Image post-processing, and radiomics analysis, including Lesion segmentation, feature extraction, Feature selection, Classification and verification, and statistical analyses. The Results section presented the Clinical characteristics, repeatability and reproducibility of T1 mapping-based radiomics features, and the performance of the models. The Discussion section interpreted these findings, considering their implications and limitations. Finally, the Conclusion section summarized the key points and suggested potential areas for future research.

Patients and methods

Patients

The retrospective study was approved by the institutional review board of The First people’s Hospital of Yancheng (No. 2022-k-115) and a waiver of informed consent was remitted. The inclusion criteria were as follows: (1) patients suspected of having lung cancer by CT with maximum diameters of tumours > 1 cm and volumes of solid component greater than half the volume of the lesion; (2) no antitumor therapies and invasive examinations before MR examination; (3) absense of MR examination contraindications; and (4) confirmed lung cancer by pathology within one month. The exclusion criteria were as follows: (1) MR images lacking T1 mapping images or unsatisfactory image quality that demonstrated serious artefacts (n = 12); and (2) patients lacking general clinical information, including age, sex and smoking history (n = 5). Finally, a total of 112 consecutive patients (75 men and 37 women; age range, 32–82 years; mean age, 64.9 years) were included in this study from June 2020 to October 2022.

Classification of the tumors as either adenocarcinoma or other types of lung cancer was based on hemotoxylin and eosin (H&E) staining according to the World Health Organization (WHO) classification of malignant lung tumors [17] by a senior pathologist with 15 years of work experience.

MRI protocols

MRI examinations were performed on a 3T MR scanner (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) using an 18-channel body surface coil. T1 mapping used 3D variable flip angle (VFA) method with flip angles (FAs) of 3° and 5°, TR/TE = 5.01/2.3 msec, matrix: 135 × 224, FOV: 380 × 305 mm2, slice thickness: 4 mm, interslice gap: 0.8 mm, 64-slice, scanning time 20s. B1-field inhomogeneity correction sequence (T1-Turbo fast low angle shot [FLASH]) was implemented to reduce the effects of magnetic field inhomogeneity before T1 mapping (10s). Inline T1 mapping was automatically generated at the scanner using the MapIt software (Siemens Healthineers). Among the patients, 54 patients underwent the repeated scans with an interval of 30 min in order to evaluate the test-retest reproducibility of T1 mapping.

Image post-processing

T1 mapping was analyzed on a postprocessing workstation (syngoMMWP, Siemens Healthcare). Two radiologists (observers A and B, with 20 and 8 years of MR diagnostic experience, respectively) independently measured native T1 values on T1-mapping pseudocolor maps without knowing the pathological results. Regions of interest (ROIs) were manually delineated along the edge of the lung cancer lesions at the level of maximum transverse diameter and visually identified hemorrhage, necrosis, calcification, large vessels, bronchi and artifacts in the lesion were excluded by referring to CT and MR images. The mean native T1 values of lung cancer were recorded. The ROIs were drawn three times, and the averages of the three values were taken as the final measurement results.

Lesion segmentation and feature extraction

Lesion segmentation were performed using the online Deepwise Research Platform (https://keyan.deepwise.com), V1.6.2. Two independent radiologists (observers A and B, with 20 and 8 years of experience in MRI reading, respectively) who were blinded to the pathological findings used a semiautomatic segmentation method to delineate the volume of interest (VOI) slice by slice. To assess the intraobserver repeatability, the observer A delineated the VOI twice with the interval of two weeks.

Radiomics feature extraction was performed using the Python package Pyradiomics (version 3.6.11). Before feature extraction, the images underwent spacing standardization using sitkBSpline interpolation sampling in Python package SimpleITK to same pixel spacing in all three dimensions. Then wavelet, Laplacian of Gaussian (LoG), Square, SquareRoot, Logarithm, Exponential, Gradient, LBP2D and LBP3D filters were also applied to the original MR images. The radiomics features included 396 first-order features, 14 shape-based features, and 1496 texture features [including gray-level co-occurrence matrix (GLCM), gray-level size zone matrix (GLSZM), gray-level run-length matrix (GLRLM), and gray-level dependence matrix (GLDM)]. Finally, T1 mapping has a total of 1906 radiomics features were extracted from the original images and the filtered images for each lesion.

Feature selection

Firstly, we retained the T1 mapping radiomics features that met the intra- and inter-observer repeatability and test-retest reproducibility. Then, the feature data were preprocessed with feature standardization by calculating Z-score. Then the data of 112 cases were randomly divided into training cohort and test cohort with the ratio of 7:3. Finally, there were 77 and 35 cases in the training cohort and test cohorts, respectively.

Feature selection including two steps: First, the Student’s t-test (normal distribution and variance homogeneity) or the Mann-Whitney U tests (non-normal distribution or unequal variances) was used to select the features that were significantly different between the adenocarcinoma and nonadenocarcinoma groups in the training cohort. Second, the least absolute shrinkage and selection operator (LASSO) was used for feature reduction [18]. In order to avoid feature selection bias and over-fitting, 10 fold cross validation was used and repeated for 50 times.

Classification and verification

The classification model was generated based on the optimal features selected in the above procedures, and logistic regression (LR) classifier was trained in the training cohort and evaluated with the validation cohort. The classification performances of T1 mapping based radiomics model, clinical model and combined model were assessed using the area under the receiver operating characteristic (ROC) curves (AUC), and sensitivity, specificity and accuracy were calculated. When the AUC was larger than 0.75, suggesting a good classification performance; while it was smaller than 0.6, suggesting a poor classification performance [19]. The flow chart of the radiomics models for the subtype differentiation of lung cancer is shown in Fig. 1.

Fig. 1
figure 1

The flowchart of the radiomics framework for the subtype discrimination of lung cancer

Full size image

Statistical analysis

Statistical analysis of the demographics and tumour characteristics were assessed using SPSS 20.0 (IBM, Chicago, IL, USA). The differences in the patient’s age, tumour size and native T1 values in the training and validation sets as well as in the adenocarcinoma and nonadenocarcinoma cohorts were assessed using an independent sample t-test or Mann-Whitney U test, and the differences in sex, smoking history and tumour location were assessed by the χ2 test. Other statistical analysis were conducted using R software (version, 4.0.3, http://r-project.org). ICCs were calculated to assess the intra- and interobserver repeatability and test-retest reproducibility of the radiomics features. Stable features that meet criteria of reproducibility were selected using the threshold of ICC > 0.80 [16, 20]. Delong test was used to compare classification performances of the radiomics model, clinical model and combined model. P < 0.05 was considered statistically significant.

Results

Clinical characteristics

As a total of 112 lesions, there were 54 cases of adenocarcinoma, 27 cases of squamous cell carcinoma, 28 cases of small cell lung cancer (SCLC) and 3 cases of carcinoid tumors. Among them, 41 cases were confirmed by surgery, 44 cases were confirmed by percutaneous puncture, 23 cases confirmed by were bronchoscopy, and 3 cases were confirmed by pleural fluid cytology. The lesions were divided into two groups: 77 cases in the training cohort and 35 cases in the validation cohort. The baseline demographic, tumour characteristics and native T1 values in the training and validation cohorts were not significantly different (P = 0.118–0.895) (shown in Table 1). Among the clinical characteristics, smoking history, lesion type and native T1 values were the predictors of pathological type of lung cancer in the training group (P = 0.004–0.038), while patients’ age, sex, the size and location of the lesions were not significant different between adenocarcinoma and non-adenocarcinoma groups (P = 0.116–0.682) (Table 1). The interobserver reproducibility of the mean T1 values of lesions was good (ICC = 0.973; 95% CI: 0.961–0.982).

Table 1 The baseline demographic and tumour characteristics in the training and validation cohorts
Full size table

Intra- and interobserver agreement of T1 mapping-based radiomics features

The median intra- and interobserver ICCs of T1 mapping were 0.933 and 0.891. There were 1576 and 1440 features with ICCs greater than 0.80, accounting for 82.69% and 75.55% of the total features, respectively (Table 2). Among the features, GLSZM texture features showed the worst measurement stability, but other texture features, first-order features and shape features showed relatively good repeatability (Table 3). In total, there were 1386 T1 mapping-based features that met the standard of agreement within and between observers at the same time with ICCs > 0.80, accounting for 72.72% of all extracted features. Figure 2 shows the intra- and interobserver agreement of T1 mapping based radiomics features.

Table 2 The intra- and interobserver agreement and test-retest reproducibility of T1 mapping-based radiomics features
Full size table
Table 3 The percents of T1 mapping-based features using filtered images with test-retest ICCs > 0.80
Full size table
Fig. 2
figure 2

The intra- and interobserver agreement and test-retest reproducibility of T1 mapping radiomics features. a and b, the intra- and interobserver repeatability of T1 mapping radiomics features; c is the test-retest reproducibility of T1 mapping radiomics features. The x-axis represents all extracted features, and the y-axis is the ICCs of the features

Full size image

Test-retest reproducibility of T1 mapping-based radiomics features

The median ICC of all radiomics features extracted from T1 mapping by repeated scans was 0.848. There were 1172 features with ICCs greater than 0.80, accounting for 61.49% of the total features (Table 2; Fig. 2). Among them, shape-based features showed optimum reproducibility with a median ICC of 0.995, and 100% (14/14) of features had ICCs greater than 0.80. While first-order, GLCM and GLSZM features showed relatively poor reproducibility (median ICCs, 0.812–0.845), only slightly more than half (53.03-57.39%) of the features had ICCs greater than 0.80 (Table 3; Fig. 3).

Fig. 3
figure 3

Interclass correlation coefficient (ICC) heat map of T1 mapping radiomics features. Shape and Gradient transform features showed the excellent reproducibility. While the first order, GLCM and GLSZM features based Exponential, LBP2D, LBP3D and Logarithm filters showed relatively poor reproducibility

Full size image

The median ICC of the original features was 0.934, and 63.00% (63/100) of the features were accompanied by ICCs greater than 0.80. Among the features based on different transformations, the Gradient transform features showed the best test-retest reproducibility (ICC > 0.80, 89.53%). While the texture features with Exponential, LBP2D, LBP3D and Logarithm filters showed relatively poor test-retest reproducibility, and stable features with ICCs greater than 0.80 were less than one half of all features (38.37-49.22%) (Table 3).

Finally, 1035 features extracted from T1 mapping that met the intra- and interobserver agreement and test-retest reproducibility were included in subsequent analysis, accounting for 54.30% of all extracted features. The stable features contained 204 first-order features, 12 shape-based features, 272 GLCM features, 193 GLDM features, 220 GLRLM features and 134 GLSZM features.

Dimensionality reduction and feature selection

After the univariate analysis and LASSO, the remaining 8 T1 mapping- based features were selected according to the tuning parameters (λ) (Fig. 4). The selected features of T1 mapping were exponential_glszm_SizeZoneNonUniformity, lbp-3D-m2_glcm_Correlation, log-sigma-5-0-mm-3D_firstorder_Kurtosis, logarithm_glcm_DifferenceEntropy, logarithm_glcm_Idn, original_shape_Sphericity, square_glcm_InverseVariance and wavelet-LHL_gldm_LargeDependenceHighGrayLevelEmphasis. The optimum features and their coefficients are shown in Fig. 5.

Fig. 4
figure 4

Feature selection using the LASSO regression method according to 10-fold cross-validation. On the basis of minimumcriteria, we selected tuning parameter (λ) (a). The vertical lines represent the optimal λ values, where the LASSO regression cross-validation obtains the minimum misclassification rate. The corresponding number of selected features is 8 for T1 mapping (b)

Full size image
Fig. 5
figure 5

The optimum features of T1 mapping and its coefficients

Full size image

Construction of the radiomics model

The performance of T1 mapping-based LR classifiers in both the training and validation cohorts were good with AUCs of 0.833 and 0.843 in the training and validation cohorts, respectively (Table 4). The diagnostic efficacy of native T1 values of lung cancer in the training and validation groups was 0.657 and 0.722, respectively. The clinical model was established by smoking history and lesion type, and the AUCs were 0.692 and 0.686 in the training and validation cohorts, respectively, showing a moderate classification performance. Combined with T1 mapping radiomics model, clinical model and T1 value to establish a comprehensive model, the prediction efficiency was further improved, and the AUCs were 0.895 and 0.915 in the training and validation cohorts, respectively. Delong test showed that the diagnostic efficacy of the combined model in the validation group was significantly higher than that of the T1 value and the clinical models (z = 2.318, P = 0.033; z = 2.194, P = 0.028), but there was no significant statistical difference between the other models (P = 0.098–0.781) (Fig. 6; Table 4).

Fig. 6
figure 6

The classification performance of radiomics, T1 value and clinical models in the training (a) and validation (b) cohorts, respectively. Combination of T1 mapping radiomics model, native T1 value and clinical characteristics can improve the predictive value of pathological types of lung cancer (a and b, AUC, 0.895 and 0.915)

Full size image
Table 4 The performances of T1 mapping radiomics model, clinical model and comprehensive model
Full size table

Discussion

This study explored the feasibility of T1 mapping in predicting the pathological type of lung cancer. The results showed that the native T1 value of lung cancer and T1 mapping based logistic regression radiomics model had the potential value in differentiating adenocarcinoma from non-adenocarcinoma, indicating that T1 mapping could potentially quantify the pathological characteristics of lung cancer and it was a promising imaging technology in tumors.

In the present study, 3D VFA T1 mapping with B1 correction was used in this study, and it has been reported that this method may make the measurement of T1 more accurate and stable [21, 22]. Additionally, we used 3D and semiautomatic segmentation methods as they yield more robust imaging features and are less time-consuming [23, 24]. The results showed that the majority of features extracted from T1 mapping were stable between repeated measurements.,which were inconsistent with the study of Yan et al. [14]. It may be due to the use of manual and 2D outlining methods and inclusion of benign lung lesions in Yan et al. [14] study, as they may have irregular shapes and blurred edges. In the study, the test-retest reproducibility of T1 mapping radiomics features was slightly worse than that of T1 VIBE, TRUFISP, CT and PET-CT reported in previous lung cancer studies [25, 26], but better than the test-retest reproducibility of ADC radiomics features reported in Peerlings et al. [15] study. The inferior reproducibility of features extracted from T1 mapping and ADC images may be because the image resolution was relatively lower. Furthermore, the inconformity in radio frequency (RF) coil sensitivity, the changes in molecular motion within lesions and the breathless state during test-retest scans may be responsible for the variations [15].

In the present study, almost all shape features extracted from T1 mapping displayed optimum reproducibility, suggesting that the semi-automatic segmentation methods used in this study was stable and it may be a surrogate of tumour volume in longitudinal follow-up, consistent with previous studies[24,27.28]. The GLDM and GLRLM texture features displayed robust stability, while the GLCM and GLSZM texture features displayed relatively poor measurement reproducibility, illustrating that they may be more sensitive to changes in noise, spatial and density resolution, scanners and measurment, similar with CT texture features [23, 27, 28]. For different filters, the Gradient, LoG, SquareRoot and Wavelet transformations improved the stability, especially the Gradient transformation, presumably due to that the effects of image noise are reduced by these filters [29]. While the Exponential, LBP2D, LBP3D, Logarithm and Square transformations reduced the stability of the original features, suggesting the necessity to evaluate the repeatability in clinical application.

In the study, the performances of the T1 mapping-based LR radiomics model for identifying the histological types of lung cancer were good with AUCs of 0.83 and 0.84 in the training and validation cohorts, which was consistent with studies on differentiating benign and malignant lung lesions [14, 30]. Additionally, Wang et al. [30] reported that the classification performance of T1WI-based radiomics features is similar to that of functional imaging sequence. This study only evaluated the diagnostic performance of a single sequence, and the efficiency was slightly higher than those reported in previous studies [6, 7, 14, 30] using T1WI, T2WI or ADC for diagnosing as well as predicting the histological subtypes and grades of lung cancer (AUC, 0.75–0.82), potentially illustrating a higher clinical value of T1 mapping for identifying the pathological characteristics of lung cancer.

In this study, we also explored the effectiveness of native T1 values for predicting the pathological types of lung cancer. The results showed that the native T1 values were different between patients with adenocarcinoma and non-adenocarcinoma, similar with previous studies [12, 13]. The possible reasons may be that SCLC and squamous cell carcinoma are mostly central type of lung cancer, and it is difficult to distinguish masses from the surrounding blood vessels when outlining the ROI. Secondly, higher degrees of micronecrosis and incomplete necrosis may occur in SCLC and squamous cell carcinoma because of large size and rapid growth, which can not be recognized by naked eyes and is therefore difficult to exclude in ROI delineation. Studies of hepatocellular carcinoma and clear cell renal cell carcinoma also found that high-grade tumors, presented as a high degree of malignancy, have higher native T1 values [31, 32]. However, in this study, the native T1 values are inferior to T1 mapping based radiomics model in the differential diagnosis of lung cancer. In Kim et al. [9] and Jensen et al. [33] studies, texture analysis of T1WI showed a better performance for discriminating pulmonary lymphoma and fungal pneumonia than T1 relaxation times and signal intensity quotients, which was similar to the results of this study.

Studies [5,6,7,8] have also reported that the combination of MR sequences and construction of a clinical-radiomics nomogram further improves the classification performance. This study haven’t included T2WI, DWI and other scanning sequences, but we combined the patient’s smoking history, lesion type (peripheral or central) and the native T1 values, and the model showed an excellent classification performance.

This study constructed an LR machine-learning model, and other machine-learning models need to be compared, such as, Random Forest (RF) and Support Vector Machine (SVM) [34]. Meanwhile, as shown in this study, nearly 50% of the T1 mapping-based radiomics features displayed relatively poor repeatability and reproducibility. Deep-learning can automatically quantify and select the most robust features to learn semantic information more effectively. The combination of deep learning features and machine-learning methods can improve the exactness and the repeatability of anticipating immunotherapy adequacy in lung cancer [35]. Meanwhile, deep learning-based methods are increasingly being used to generate and segmentate images [36, 37]. Multimodal fusion artificial intelligence diagnosis system increases the diagnostic efficiency and accuracy [38], which may be the leading direction of future development.

This study has several limitations. First, this is a single-centre study and the sample size was relatively small, and further multi-center and large sample studies are needed to improve the robustness of the model. Second, the study was retrospective, and there were more male patients than female patients, resulting in potential selection bias. Third, this study included SCLC patients, which may exist confounding factors, but it made the clinical application of this radiomics model wider. Fourth, due to the incomplete archival database, this study excluded other potential clinical features, such as carcinoembryonic antigen (CEA) and genetic mutations, which require further analysis. Fifth, this study only explored the fusion model of T1 mapping sequence with clinical and imaging features. Other sequences, such as T1WI, T2WI, and ADC, need to be integrated for further research. Moreover, the comparation with other machine-learning and deep-learning models are still needed in further studies.

Conclusion

In conclusion, our study demonstrated the feasibility of T1 mapping-based logistic regression radiomics model in the identification of histological types of lung cancer, while about 50% of the T1 mapping-based radiomics features displayed relatively poor repeatability and reproducibility, so the reproducibility of radiomics features needs to be evaluated when building a radiomics model. The combination of T1 mapping radiomics model, clinical characteristics and native T1 value can improve the predictive value of pathological types of lung cancer, potentially implying that T1 mapping is a rapid, promising and non-radiation imaging technology for characterizing lung cancer. Additionally, it can be clinically useful in patients with lung cancer and extended to the identification of other diseases. In the future, T1 mapping may be able to be incorporated into multimodal artificial intelligence diagnostic systems to make the disease diagnosis more efficient and accurate.

Data availability

All data generated or analysed during this study are included in the published article, further inquiries can be directed to the corresponding auther on reasonable request.

Abbreviations

ADC:

Apparent Diffusion Coefficient

AUC:

Area Under Curve

CEA:

Carcinoembryonic Antigen

CT:

Computed Tomography

DWI:

Diffusion-Weighted Imaging

GRE:

Gradient-Recalled Echo

ICC:

Intra-Class Correlation Coefficient

LASSO:

Least Absolute Shrinkage and Selection Operator

LR:

Logistic Regression

MRI:

Magnetic Resonance Imaging

ROC:

Receiver Operating Characteristic

SCLC:

Small Cell Lung Cancer

T1WI:

T1-Weighted Imaging

T2WI:

T2-Weighted Imaging

VFA:

Variable Flip Angle

VIBE:

Volume Interpolated Breath Hold Examination

VOI:

Volume Of Interest

WHO:

World Health Organization

References

  1. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71(1):7–33.

    Article  PubMed  Google Scholar 

  2. Ettinger DS, Wood DE, Aisner DL, et al. Non-small cell lung cancer, Version 5.2017, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw. 2017;15(4):504–35.

    Article  PubMed  Google Scholar 

  3. Yue JY, Chen J, Zhou FM, et al. CT-pathologic correlation in lung adenocarcinoma and squamous cell carcinoma. Med (Baltim). 2018;97(50):e13362.

    Article  Google Scholar 

  4. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Xie K, Cui C, Li X et al. MRI-based clinical-imaging-radiomics nomogram model for discriminating between benign and malignant solid pulmonary nodules or masses. Acad Radiol. 20246 – 6332(24)00207-1. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.acra.2024.03.042. Online ahead of print.

  6. Tang X, Bai G, Wang H, et al. Elaboration of multiparametric MRI- based radiomics signature for the preoperative quantitative identification of the histological grade in patients with non-small-cell lung cancer. J Magn Reson Imaging. 2022;56(2):579–89.

    Article  PubMed  Google Scholar 

  7. Tang X, Xu X, Han Z, et al. Elaboration of a multimodal MRI-based radiomics signature for the preoperative prediction of the histological subtype in patients with non-small-cell lung cancer[J]. Biomed Eng Online. 2020;19(1):5.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Wang Y, Wan Q, Xia X, et al. Value of radiomics model based on multi-parametric magnetic resonance imaging in predicting epidermal growth factor receptor mutation status in patients with lung adenocarcinoma[J]. J Thorac Dis. 2021;13(6):3497–508.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Kim D, Elgeti T, Penzkofer T, et al. Enhancing the differentiation of pulmonary lymphoma and fungal pneumonia in hematological patients using texture analysis in 3-T MRI. Eur Radiol. 2021;31(2):695–705.

    Article  PubMed  Google Scholar 

  10. Sibley CT, Noureldin RA, Gai N, et al. T1 mapping in cardiomyopathy at cardiac MR: comparison with endomyocardial biopsy. Radiology. 2012;265(3):724–32.

    Article  PubMed  PubMed Central  Google Scholar 

  11. Yang S, Shan F, Yan Q, et al. A pilot study of native T1-mapping for focal pulmonary lesions in 3.0 T magnetic resonance imaging: size estimation and differential diagnosis. J Thorac Dis. 2020;12(5):2517–28.

    Article  PubMed  PubMed Central  Google Scholar 

  12. Li G, Huang R, Zhu M, et al. Native T1-mapping and diffusion-weighted imaging (DWI) can be used to identify lung cancer pathological types and their correlation with Ki-67 expression. J Thorac Dis. 2022;14(2):443–54.

    Article  PubMed  PubMed Central  Google Scholar 

  13. Jiang J, Cui L, Xiao Y, et al. B(1) -Corrected T1 mapping in lung cancer: repeatability, reproducibility, and identification of histological types. J Magn Reson Imaging. 2021;54(5):1529–40.

    Article  PubMed  Google Scholar 

  14. Yan Q, Yi Y, Shen J, et al. Preliminary study of 3 T-MRI native T1-mapping radiomics in differential diagnosis of non-calcified solid pulmonary nodules/masses. Cancer Cell Int. 2021;21(1):539.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Peerlings J, Woodruff HC, Winfield JM, et al. Stability of radiomics features in apparent diffusion coefficient maps from a multi-centre test-retest trial. Sci Rep. 2019;9(1):4800.

    Article  PubMed  PubMed Central  Google Scholar 

  16. Merisaari H, Taimen P, Shiradkar R, et al. Repeatability of radiomics and machine learning for DWI: short-term repeatability study of 112 patients with prostate cancer. Magn Reson Med. 2020;83(6):2293–309.

    Article  PubMed  Google Scholar 

  17. Travis WD, Brambilla E, Nicholson AG, et al. The 2015 World Health Organization classification of lung tumors: impact of genetic, clinical and radiologic advances since the 2004 classification. J Thorac Oncol. 2015;10(9):1243–60.

    Article  PubMed  Google Scholar 

  18. Chi MA, Huang J, Mathematics D et al. Asymptotic properties of Lasso in high-dimensional partially linear models. Sci China(Mathematics) 2016.

  19. Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA. 2017;318(14):1377–84.

    Article  PubMed  Google Scholar 

  20. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull. 1979;86(2):420–8.

    Article  CAS  PubMed  Google Scholar 

  21. Cheng HL, Wright GA. Rapid high-resolution T(1) mapping by variable flip angles: accurate and precise measurements in the presence of radiofrequency field inhomogeneity. Magn Reson Med. 2006;55(3):566–74.

    Article  PubMed  Google Scholar 

  22. Yoon JH, Lee JM, Kim E, et al. Quantitative liver function analysis: volumetric T1 mapping with fast multisection B(1) inhomogeneity correction in hepatocyte-specific contrast-enhanced liver MR imaging. Radiology. 2017;282(2):408–17.

    Article  PubMed  Google Scholar 

  23. Zhao B, Tan Y, Tsai WY, et al. Reproducibility of radiomics for deciphering tumor phenotype with imaging. Sci Rep. 2016;6:23428.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  24. Parmar C, Rios Velazquez E, Leijenaar R, et al. Robust radiomics feature quantification using semiautomatic volumetric segmentation. PLoS ONE. 2014;9(7):e102107.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Mahon RN, Hugo GD, Weiss E. Repeatability of texture features derived from magnetic resonance and computed tomography imaging and use in predictive models for non-small cell lung cancer outcome. Phys Med Biol 2019.

  26. Leijenaar RT, Carvalho S, Velazquez ER, et al. Stability of FDG-PET Radiomics features: an integrated analysis of test-retest and inter-observer variability. Acta Oncol. 2013;52(7):1391–7.

    Article  CAS  PubMed  Google Scholar 

  27. Reiazi R, Abbas E, Famiyeh P, et al. The impact of the variation of imaging parameters on the robustness of computed tomography radiomic features: a review. Comput Biol Med. 2021;133:104400.

    Article  PubMed  Google Scholar 

  28. Choe J, Lee SM, Do KH, et al. Deep learning-based image conversion of CT reconstruction kernels improves radiomics reproducibility for pulmonary nodules or masses. Radiology. 2019;292(2):365–73.

    Article  PubMed  Google Scholar 

  29. Yasaka K, Akai H, Mackin D, et al. Precision of quantitative computed tomography texture analysis using image filtering: a phantom study for scanner variability. Med (Baltim). 2017;96(21):e6993.

    Article  Google Scholar 

  30. Wang X, Wan Q, Chen H, et al. Classification of pulmonary lesion based on multiparametric MRI: utility of radiomics and comparison of machine learning methods. Eur Radiol. 2020;30(8):4595–605.

    Article  PubMed  Google Scholar 

  31. Wang S, Li J, Zhu D, et al. Contrast-enhanced magnetic resonance (MR) T1 mapping with low-dose gadolinium-diethylenetriamine pentaacetic acid (Gd-DTPA) is promising in identifying clear cell renal cell carcinoma histopathological grade and differentiating fat-poor angiomyolipoma. Quant Imaging Med Surg. 2020;10(5):988–98.

    Article  PubMed  PubMed Central  Google Scholar 

  32. Peng Z, Jiang M, Cai H, et al. Gd-EOB-DTPA-enhanced magnetic resonance imaging combined with T1 mapping predicts the degree of differentiation in hepatocellular carcinoma. BMC Cancer. 2016;16:625.

    Article  PubMed  PubMed Central  Google Scholar 

  33. Jensen LJ, Kim D, Elgeti T, et al. Differentiation of pulmonary lymphoma manifestations and nonlymphoma infiltrates in possible invasive fungal disease using fast T1-weighted magnetic resonance imaging at 3 T comparison of texture analysis, mapping, and signal intensity quotients. J Thorac Imaging. 2022;37(2):80–9.

    Article  PubMed  Google Scholar 

  34. Singh LK, Khanna M, Thawkar S et al. Nature-inspired computing and machine learning based classification approach for glaucoma in retinal fundus images. Multimedia Tools Appl 2023;1–49.

  35. Ren Q, Xiong F, Zhu P, et al. Assessing the robustness of radiomics/deep learning approach in the identification of efficacy of anti–PD-1 treatment in advanced or metastatic non-small cell lung carcinoma patients. Front Oncol. 2022;12:952749.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Ansari MY, Mangalote IAC, Meher PK, et al. Advancements in deep learning for B-Mode ultrasound segmentation: a comprehensive review. IEEE Trans Emerg Top Comput Intell. 2024;8. https://doiorg.publicaciones.saludcastillayleon.es/10.1109/TETCI.2024.3377676. Online ahead of print.

  37. Ansari MY, Qaraqe M, Righetti R, et al. Unveiling the future of breast cancer assessment: a critical review on generative adversarial networks in elastography ultrasound. Front Oncol. 2023;11(13):1282536.

    Article  Google Scholar 

  38. Singh LK, Khanna M, Garg H. Multimodal biometric based on fusion of ridge features with minutiae features and face features. Int J Inform Syst Model Des (IJISMD). 2020;11(1):37–57.

    Article  Google Scholar 

Download references

Acknowledgements

None.

Funding

This work was supported by the 2022 Yancheng Key R&D Plan Guiding Project (YCBE202211); and the Xuzhou Medical University Affiliated Hospital Science and Technology Development Excellent Talent Fund Project (XYFY202304).

Author information

Authors and Affiliations

  1. Department of Radiology, The Yancheng Clinical College of Xuzhou Medical University and The First people’s Hospital of Yancheng, Yulong West Road No. 166, Yancheng, 224001, China

    Jianqin Jiang, Yong Xiao, Gaofeng Xu & Yigang Fu

  2. Department of Radiology, The First Affiliated Hospital of Soochow University, Shizi Street No. 188, Suzhou, 215002, China

    Jianqin Jiang & Chunhong Hu

  3. Department of Radiology, The Second Affiliated Hospital of Nantong University, Haierxiang North Road No. 6, Nantong, 226001, China

    Jia Liu & Lei Cui

  4. Department of Pathology, The Yancheng Clinical College of Xuzhou Medical University and The First people’s Hospital of Yancheng, Yulong West Road No. 166, Yancheng, 224001, China

    Weiwei Shao

  5. Siemens Healthineers Digital Technology (Shanghai) Co., Ltd, Haiyang West Road No. 399, Shanghai, 200000, China

    Shaowei Hao

  6. Institute of Medical Imaging, Soochow University, Shizi Street No. 1, Suzhou, 215002, China

    Chunhong Hu

Authors
  1. Jianqin Jiang
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Yong Xiao
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Jia Liu
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Lei Cui
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Weiwei Shao
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Shaowei Hao
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Gaofeng Xu
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Yigang Fu
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Chunhong Hu
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

All authors contributed to the study conception and design. Conceptualization, methodology were performed by Chunhong Hu, Yigang Fu and Lei Cui. Material preparation, data collection and analysis were performed by Jianqin Jiang, Yong Xiao, Jia Liu, Weiwei Shao and Shaowei Hao. The first draft of the manuscript was written by Jianqin Jiang and Yong Xiao. The data was verified and proofread by Lei Cui. Supervision were performed by Gaofeng Xu, Yigang Fu and Chunhong Hu.

Corresponding authors

Correspondence to Yigang Fu or Chunhong Hu.

Ethics declarations

Ethics approval and consent to participate

This was a retrospective study whose protocol was approved by the Ethics Committee of The First people’s Hospital of Yancheng (NO. 2020 K-063) and waived the need for informed consent from patients.

Consent for publication

Not applicable.

Clinical trial number

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Jiang, J., Xiao, Y., Liu, J. et al. T1 mapping-based radiomics in the identification of histological types of lung cancer: a reproducibility and feasibility study. BMC Med Imaging 24, 308 (2024). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12880-024-01487-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12880-024-01487-y

Keywords

BMC Medical Imaging

ISSN: 1471-2342

Contact us