Identification of the Pathological Types of Brain Metastasis from Lung Cancer Based on Multiparametric MRI Radiomics: A Feasibility Study

doi:10.21203/rs.3.rs-4894127/v1

Download PDF

Research Article

Identification of the Pathological Types of Brain Metastasis from Lung Cancer Based on Multiparametric MRI Radiomics: A Feasibility Study

https://doi.org/10.21203/rs.3.rs-4894127/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Objectives

To differentiate brain metastases (BMs) from non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC) and BMs due to the adenocarcinoma (AD) and non-adenocarcinoma (NAD) subtypes using radiomic features derived from multiparametric magnetic resonance imaging (MRI).

Methods

276 patients with BMs, including 98 with SCLC and 178 with NSCLC, were randomly divided into training (193 cases) and validation (83 cases) sets in a ratio of 7:3. Of the 178 patients with NSCLC, 155 were from primary AD and 23 from NAD. These were also randomly divided into training (124 cases) and validation (54 cases) sets. A logistic regression analysis was used to construct classification models based on radiomics features that were extracted from T1 weighted contrast-enhanced (T1CE), fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted imaging (DWI) images. The receiver operating characteristic curve (ROC) was used to evaluate the diagnostic efficiency.

Results

Multiparametric combined-sequence MRI radiomics features based on TICE, FLAIR, and DWI images were highly specific in distinguishing brain metastases originating from different types of lung cancers. In the training and validation sets, the area under the curves (AUCs) of the model for the classification of SCLC and NSCLC brain metastasis were 0.765 (95% CI 0.711, 0.822) and 0.762 (95% CI 0.671, 0.845), respectively; the AUC values of the prediction models combining the three sequences in differentiating AD from NAD BMs were 0.861 (95% CI 0.756, 0.951) and 0.851 (95% CI 0.649, 0.984), respectively.

Conclusion

The radiomics classification method based on the combination of multiple MRI sequences may be used for differentiating between the various lung cancer BMs.

Lung cancer

brain metastases

radiomics

magnetic resonance imaging

pathology

Brain metastases (BMs) are the most common disseminated intracranial tumors in patients with cancer and their incidence is reported to be increasing. They cause significant morbidity and mortality due to their heterogeneous features and diagnosis at a late stage^[1]. Lung cancer metastases are the most common BMs in adults, comprising nearly 50% of BMs^[2]. The most common lung-cancer type is represented by non-small cell lung cancer (NSCLC) accounting for approximately 90%, the remaining being small cell lung cancer (SCLC)^[3]. With NSCLC, the leading histological subtype is adenocarcinoma (AD)^[4]. In the light of the different treatment options available for patients with BMs from lung cancer, a reliable classification of different histological types is highly desirable to allow early and individualized management^[5]. An invasive biopsy and histological workup are generally required for an accurate diagnosis. However, contraindications to surgical lung resection and the limitations of doing biopsies of the BMs make diagnosis a challenge for clinicians. Furthermore, most BMs are diagnosed by magnetic resonance imaging (MRI) scans^[6]. Hence, it would be of great clinical importance if the pathological type of lung cancer could be inferred through the MRI findings of the BMs.

Radiomics, a quantitative approach to MRI imaging, may provide biological, prognostic, and predictive information that is not revealed upon visual inspection of the images alone. Radiomics of the BM images holds great promise for the correlation of tumor histopathological type with the morphologic heterogeneity in MR imaging^[7]. For BMs with lung cancer, several studies have demonstrated the ability of radiomic features based on multiparametric MRI to differentiate BMs from various primary cancers^[8] and to predict the genetic mutation status^{[9, 10]}. Prior studies have substantiated the performance of radiomic methods in patients with primary lung cancer to both visualize and quantify histopathological changes before treatment^{[11, 12]}. However, most studies in pathological classification have relied solely on the primary lung cancer^{[13, 14]}. Thus, a non-invasive imaging-focused radiomic approach to predict pathological subtypes from BMs would complement the radiomics information available on the primary lung cancer. The aim of the current study was to determine whether multiparametric MRI can help to differentiate the various pathological types of BMs including SCLC, NSCLC, AD, and non-adenocarcinoma (NAD).

Subjects

This retrospective study was approved by the ethics committee of our hospital. Written informed consent to participate was obtained from all patients or their family members. Between January 2016 and June 2019, patients with BMs from lung cancers who had the MRI transverse T1-weighted contrast-enhanced (T1CE), fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted imaging (DWI) data in our hospital were enrolled. The primary lung cancer was diagnosed on the basis of histopathology, and this was taken as the gold standard for diagnosis. All BMs were confirmed on MRI by two radiologists. A biopsy confirmed the diagnosis of primary NSCLCs and the histological differentiation into either AD or NAD. The inclusion criteria were patients with primary lung cancers with concomitant BMs confirmed via biopsy, surgery, or imaging. The exclusion criteria were patients with co-existence of other malignant tumors in the lung or brain, poor quality of imaging for analysis, and a history of craniocerebral surgery.

MR images acquisition

In this study, all patients underwent a brain MRI at one of the three MRI scanners (Siemens Amira 1.5T, Philips Achieva 1.5T, or GE Discovery MR750 3.0T) housed at our hospital. Gadodiamide was used as a contrast agent for cranial MRI. A high-pressure syringe was used to inject 0.01 mmol/kg (0.2 mL/kg) of the contrast through the cubital vein at an injection rate of 2 mL/s.

The typical scanning parameters for a 3 Tesla (T) machine scan were as follows:

Transverse T1CE: repetition time (TR)/echo time (TE) (ms) = 2320/80, number of signal averages (NSA) = 1, field of view (FOV) = 21 cm × 32 cm, thickness/gap (mm) = 5.0/1.0.
Transverse FLAIR: TR/TE (ms) = 8000/109, NSA = 1, FOV = 24 cm × 36 cm, thickness/gap (mm) = 5.0/1.0.
Transverse DWI: TR/TE (ms) = 1,306/54, NSA = 2, FOV = 24 cm × 36 cm, thickness/gap (mm) = 5.0/1.0, B value selected 800 s/mm².

Tumor segmentation and features extraction

The images were exported into a Digital Imaging and Communications in Medicine (DICOM) format, and the dataset was randomly assigned in a 7:3 ratio to either the training cohort or the test cohort. All cases in the training cohort were used to train the predictive models, while cases in the test cohorts were used to independently evaluate the performance of the models. Before analyses, the whole volume of the region of interest (ROI) of the tumor was manually contoured by two radiologists on T1CE, FLAIR, and DWI images using ITK-SNAP (Version 3.8.0, www.itksnap.org) software.As shown in Fig. 2, figures (A2, B2, C2, D2, E2, F2, G2, H2, I2) show the ROI range of patients with SCLC, AD, and NAD with BMs.

Then the original images were standardized by the z-score standardized method, and the gray value of the images was adjusted to a standard normal distribution. Finally, the pre-processed images and corresponding contoured target ROIs were imported into the artificial intelligence software (Version 3.3.0, Artificial Intelligence Kit, GE Healthcare, China) for the extraction of texture features from five categories of feature parameters, including histogram, form factor, gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), and gray-level size zone matrix (GLSZM). Finally, 1,037 features for each sequence of each case were extracted, giving a total of 3,111 radiomics features from each patient.

Radiomics features selection and model building

To prevent the problem of overfitting, radiomic features need to be screened. The selection of features was performed using univariate logistic analysis followed by step-wise multivariate logistic analysis. Firstly, the features with a P value < 0.05 from the univariate logistic analysis were retained (Mann-Whitney U test). Subsequently, using the Spearman correlation coefficients generated by correlation analysis, features with characteristic parameters above 0.9 were eliminated. Finally, the multivariable logistic regression was used to determine the optimal features using the Akaike information criterion (AIC) as the threshold. Images from three MRI sequences (T1CE, DWI, FLAIR) were used for building corresponding prediction logistic models based on optimal radiomics parameters, namely, the T1CE model, the DWI model, the FLAIR model, and the multi-sequence model.

Statistical analysis

The clinical data of enrolled patients were analyzed using the R Statistical Software (Version 3.6.3 available at www.rproject.org). The differences between the two groups were compared by independent sample t-test or Mann-Whitney U test. An receiver operating characteristic curve (ROC) curve analysis was performed to determine the performance of each logistic model, and the area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) were calculated. Statistical analyses for the MRI parameters were performed using the packages R version 3.5.1 and Python version 3.5.6. A two-tailed P-value < 0.05 indicated statistical significance.

Patients’ characteristics

The flowchart for patient selection is shown in Fig. 1. A total of 276 BMs patients admitted between June 2016 to June 2021 were reviewed. Of these, 178 were of primary NSCLC and 98 of SCLC. Of the 178 NSCLC patients, 155 were AD and 23 were NAD. The 276 BMs patients were randomly divided into training (70%) and testing cohorts (30%), respectively. Table 1 shows the clinical characteristics of the enrolled patients and their division into training and testing data sets classified according to the individual tumors. There were no statistically significant differences in sex age, smoking and stage of lung cancer in the training and testing groups.

Table 1

characteristics of enrolled patients in the training and testing data sets
Characteristics	Training data set(n = 193) SCLC NSCLC P value (n = 124) (n = 69)	Testing data set(n = 83) SCLC NSCLC P value (n = 54) (n = 29)	Training data set(n = 124) AD NAD P value (n = 108) (n = 16)	Testing data set(n = 54) AD NAD P value (n = 47) (n = 7)
Gender(n) Male Female Age (years) mean ± std Smoking No Yes Stage M1b M1c	0.333 80 47 44 22 0.137 59.9 ± 10.15 63.7 ± 7.54 0.196 49 38 75 31 0.145 25 40 99 29	0.341 29 22 25 7 0.267 62.7 ± 7.46 57.0 ± 13.04 0.298 25 17 29 12 0.385 22 15 32 14	0.234 57 15 51 1 0.247 60.4 ± 9.50 62.4 ± 12.44 0.279 66 10 42 6 0.218 68 9 40 7	0.239 30 7 17 0 0.331 60.8 ± 8.83 61.6 ± 4.17 0.142 29 5 18 2 0.218 27 3 20 4
Note: SCLC, small-cell lung cancer; NSCLC, non-small-cell lung cancer; MR, magnetic resonance; AD, adenocarcinoma; NAD, non-adenocarcinoma.

Classification of performance of the models

Performance of models to differentiate NSCLC from SCLC

Four, three, four, and eleven radiomics features, respectively, were extracted from T1CE, FLAIR, DWI, and combined sequences after screening. The optimal features for the classification of NSCLC and SCLC were selected from the combined sequential MRI using univariate logistic analysis and multivariate logistic analysis (Table 2).

Table 2

List of selected radiomics features
Number	Features	Senquences	Coefficient
1 2 3 4 5 6 7 8 9 10 11	original_shape_SurfaceVolumeRatio wavelet-HLL_glcm_Correlation log-sigma-3-0-mm-3D_firstorder_Skewness wavelet-LLH_glszm_LowGrayLevelZoneEmphasis wavelet-HHH_glszm_SmallAreaLowGrayLevelEmphasis wavelet-HLL_firstorder_Skewness wavelet-HHL_glrlm_RunEntropy wavelet-LLL_glcm_Autocorrelation wavelet-HLH_glcm_MaximumProbability original_gldm_SmallDependenceHighGrayLevelEmphasis log-sigma-3-0-mm-3D_ngtdm_Busyness	T1CE T1CE T1CE T1CE FLAIR FLAIR FLAIR DWI DWI DWI DWI	0.0395 -0.3550 -0.4296 0.2481 -0.4599 0.2140 -0.3319 -1.0478 0.2784 0.8169 0.6544

The performance of the three single-sequence models was not ideal, and the AUCs in the training set and the validation set were all lower than 0.700, except for SPE and PPV. The multi-sequence model achieved better performance when compared to the single-sequence models; the AUC, ACC, SEN, SPE, PPV and NPV in the training set were 0.765 (95% confidence intervals [CI]: 0.711, 0.822), 0.694, 0.694, 0.797, 0.849, and 0.550, respectively; and in the test group, the AUC, ACC, SEN, SPE, PPV, and NPV were 0.762 (95% CI: 0.671, 0.845), 0.675, 0.648, 0.724, 0.814, and 0.525, respectively. Classification results of the proposed radiomic models for differentiation between NSCLC from SCLC are given in Table 3 and the ROC values are shown in Fig. 3. The calibration curve for the probability of identifying NSCLC and SCLC in the training and the validation groups demonstrated good agreement between prediction and observation (Figure. 3).

Table 3

Model performance comparison for the three individual and combined MR sequences
Models	Training data set T1CE FLAIR DWI T1CE + DWI + FLAIR		Testing data set T1CE FLAIR DWI T1CE + DWI + FLAIR
AUC(95%CI) ACC SEN SPE PPV NPV	0.657 (0.591,0.719) 0.676 (0.608,0.743) 0.570 0.570 0.407 0.570 0.899 0.855 0.873 0.836 0.449 0.447	0.686 (0.622,0.747) 0.765 (0.711,0.822) 0.539 0.694 0.315 0.694 0.942 0.797 0.907 0.849 0.433 0.550	0.632 (0.524,0.738) 0.652 (0.547,0.748) 0.667 (0.564,0.766) 0.762 (0.671,0.845) 0.530 0.627 0.506 0.675 0.407 0.627 0.296 0.648 0.759 0.862 0.897 0.724 0.759 0.871 0.842 0.814 0.407 0.481 0.406 0.525
Note: T1CE, contrast-enhanced T1 weighted imaging; FLAIR, fluid attenuated inversion recovery; DWI, diffusion-weighted imaging.

Performance of models to differentiate AD from NAD

Three, two, three, and eight radiomics features, respectively, were extracted from T1CE, FLAIR, DWI, and combined sequences after screening. The optimal features for differentiating AD and NAD were selected from the combined sequential MRI using univariate logistic analysis and multivariate logistic analysis (Table 4).

Table 4

List of selected radiomics features
Number	Features	Senquences	Coefficient
1 2 3 4 5 6 7 8	wavelet-HLH_glcm_InverseVariance log-sigma-5-0-mm-3D_gldm_DependenceNonUniformity wavelet-HHH_gldm_DependenceEntropy wavelet-HHH_gldm_DependenceEntropy wavelet-HLH_glszm_LargeAreaLowGrayLevelEmphasis waveletHHH_gldm_LargeDependenceHighGrayLevelEmphasiswavelet-HLL_ngtdm_Busyness original_glcm_ClusterShade	T1CE T1CE T1CE FLAIR FLAIR DWI DWI DWI	0.8599 0.1958 -0.1455 -0.9417 0.7466 0.9760 0.5718 -0.6176

Among the three single-sequence models, the T1CE model had the best prediction performance. The AUCs of the training group and the test group in the T1CE model were 0.787 (95% CI 0.676,0.882) and 0.772 (95% CI 0.558, 0.944), respectively. Compared to the T1CE model, the prediction efficiency of the FLAIR and DWI models were relatively lower, with AUCs equaling 0.768 (95% CI 0.645, 0.877) and 0.741 (95% CI 0.635, 0.833) in the training group and 0.763 (95% CI 0.618, 0.900) and 0.733 (95% CI 0.468, 0.984) in the test group. However, compared with the single sequence model, the multi-sequence model achieved even better performance; the AUC, ACC, SEN, SPE, PPV, and NPV in the training set were 0.861 (95% CI: 0.756, 0.951), 0.821, 0.750, 0.832, 0.400, and 0.957, respectively; and in the test group, the AUC, ACC, SEN, SPE, PPV, and NPV were 0.851 (95% CI 0.649, 0.984), 0.871, 0.857, 0.872, 0.500, and 0.976, respectively. Classification results of the proposed logistic models for differentiation between AD and NAD are given in Table 5 and ROC values are shown in Fig. 5. The calibration curve for the probability of identifying AD from NAD in the training and the validation groups demonstrated good agreement between prediction and observation (Figure. 6).

Table 5

Model performance comparison for the three individual and combined MR sequences
Models	Training data set T1CE FLAIR DWI T1CE + DWI + FLAIR		Testing data set T1CE FLAIR DWI T1CE + DWI + FLAIR
AUC(95%CI) ACC SEN SPE PPV NPV	0.787 (0.676,0.882) 0.768 (0.645, 0.877) 0.691 0.732 0.821 0.625 0.776 0.748 0.314 0.270 0.943 0.930	0.741 (0.635, 0.833) 0.861 (0.756, 0.951) 0.642 0.821 0.750 0.750 0.626 0.832 0.231 0.400 0.944 0.957	0.772 (0.558,0.944) 0.763 (0.618,0.900) 0.733(0.468,0.984) 0.851 (0.649,0.984) 0.723 0.685 0.593 0.871 0.714 0.732 0.732 0.857 0.745 0.681 0.574 0.872 0.278 0.250 0.200 0.500 0.944 0.941 0.931 0.976
Note: T1CE, contrast-enhanced T1 weighted imaging; FLAIR, fluid attenuated inversion recovery; DWI, diffusion-weighted imaging.

An ability to type the primary lung cancer from the MRI images of the BMs would be of great advantage to clinicians especially when conventional invasive procedures like biopsy are difficult to perform^[15]. Our results showed that the MR imaging-based radiomic analysis of BMs can serve as a noninvasive technique to differentiate the primary SCLC from NSCLC, and AD from NAD in lung cancer patients. Radiomics holds immense potential for the correlation of imaging features with histopathological changes. These are difficult to achieve by visual inspection alone^[16]. Our study provides evidence of the role of radiomics in complementing traditional methods for the pathological classification of BMs.

Our study was novel in that we analyzed MR images of BMs using a radiomics method to classify different lung cancer subtypes. Previous classification of lung cancer by radiomics approach has mainly focused on primary tumors. In a previous study, Tang et al^[17] presented the potential of multimodal chest MRI-based radiomics for discrimination between lung squamous cell carcinoma and AD subtypes of NSCLC. Their model obtained an AUC of 0.824 for the test set, which was comparable to computed tomography (CT) or positron emission tomography (PET) based radiomic models for the identification of NSCLC subtypes^{[18, 19]}. Wang et al^[20] demonstrated that the multi-parametric MR model has a better predictive efficiency compared with CT and PET-bases radiomic models.

In our study, the multi-sequence radiomics model performed better as compared with the single-sequence models. Studies showed that DWI was of great value in the evaluation of BMs and have demonstrated that most BMs of lung cancer show hyperintensity on DWI^[21]. Furthermore, the current guidelines^[22] also recommend T1CE as an effective tool and an important indicator in the diagnosis of BMs ,repesenting destruction of the blood-brain barrier; moreover, T2-weighted FLAIR can best detect peritumoral edema in the brain,which aids in distinguishing edema in the brain and other non-tumor-related abnormal regions during tumor evaluation.Therefore, we combined DWI with T1CE and to classify the imaging characteristics of such lesions, because it had better predictive efficiency and provided more comprehensive information about tumor characteristics.It had better predictive efficiency and provided more comprehensive information about tumor characteristics. This is consistent with a previous MR radiomics study on lung carcinoma classification^[23] that reported that the best classification results were obtained for the multiparametric MRI data, with a mean AUC of 0.90 for classification between SCLC BMs and NSCLC BMs.Similarly, Li et al^[24]conducted a multicenter study evaluating the feasibility of a deep learning approach based on multiparametric MRI to differentiate pathological subtypes of BMs in lung cancer patients,achieving AUC values of 0.796 and 0.751 in distinguishing SCLC from NSCLC BMs and differentiating AD from squamous cell carcinoma BMs.Tulum et al^[25] successfully differentiated BMs from SCLC and NSCLC (AD and squamous cell carcinoma) in small datasets (74 patients) by introducing novel radiomic features with the sensitivity and specificity values of 94.44% and 95.33%,and deep learning algorithms with the values of 94.29% and 94.08%, basing on T2 weighted and FLAIR axial images, demonstrating the efficiency and importance of these methods in aiding the classification of lung cancer BMs, and indicating that the proposed novel radiomics feature algorithm had significant advantages over deep learning algorithms in identifying lung cancer subtypes in small datasets and exhibited strong interpretability. Compared to previous studies, our research capitalizes on a larger dataset utilizing multiparametric MRI radiomics, achieving AUCs of 0.762 and 0.861 in distinguishing SCLC from NSCLC and AD from NAD BMs, respectively. Highlighting the advantages of multiparametric MRI in this context, our approach offers superior clinical visualization and interpretability compared to deep learning models^[26].

Previous studies have reported that models generated from radiomics provide promising results in differentiating pathological types of lung cancers based on BMs imaging^[27]. Zhang et al^[11] demonstrated the feasibility and accuracy of radiomics features extracted from brain-enhanced CT to identify the pathological subtypes of the primary site in the lung, achieving an AUC of 0.828 in the differentiation of primary lung adenocarcinoma and squamous cell carcinoma for patients with BMs. Although both MRI and CT are accepted as the primary modalities of screening for BMs, CT is markedly less sensitive than MRI and should be limited to patients with contraindications for MRI^[28]. Li et al^[12] reported that CT radiomics based lesion classification was highly specific in differentiating BMs of lung cancers, with misclassification rates of 3.1%, 4.3%, 5.8%, and 8.1%, for SCLC, squamous cell carcinoma, AD, and large cell lung carcinoma, respectively. They demonstrated that it was possible to differentiate BMs from SCLC and NSCLC using radiomics features extracted from T1CE images of brain metastatic lesions to predict the pathological subtypes of the originating lung cancers. This was consistent with our study, in which the predictive efficacy of the T1CE models was slightly better than that of the FLAIR and DWI models, possibly because T1CE is particularly useful for demonstrating vascularity within lesions of BMs^[29]. Similarly, Ahn et al^[30] proposed that T1CE image radiomics of BMs predicted the epidermal growth factor receptor (EGFR) mutation status of lung cancer BMs with good diagnostic performance, reaching an AUC of 89.09%. Thus, T1CE imaging has the potential to differentiate the pathological structure of tumors.T1CE not only facilitates the identification of lesion contours in BMs but also provides physicians with crucial information about lesion biology and treatment response^[31]. Thus, MRI has the potential to differentiate the pathological structure of tumors and predict the pathological types of BMs in lung cancer^[32].

There are some limitations in the current study. Firstly, this was a single-center study, with a relatively small sample size. Further, manual lesion segmentation may affect the generalizability of the model. To further improve classification efficacy and to assess the resulting clinical impact, multicenter studies are needed with standardized imaging protocols, standard post-processing procedures, and automatic tumor segmentation. Moreover, other subtypes of NSCLC, such as squamous cell carcinoma, adenosquamous and large cell lung carcinoma were excluded due to the limited number of cases available. The addition of basic clinical information (sex and age) may also help in improving the performance of the classifier models.

In conclusion, the proposed multiparametric MR radiomics models could capture critical imaging features of BMs from patients with lung cancer and differentiate the common pathological subtypes of primary lung cancer. Our study provides a complementary method based on radiomics focused on MRI imaging of BMs for the accurate prediction of pathological subtypes of lung cancer and has the potential to be used as a reference for personalized, targeted therapy of lung cancer cases.

SCLC (small cell lung cancer)

NSCLC (non-small cell lung cancer)

BMs (brain metastases )

AD(adenocarcinoma)

NAD (non-adenocarcinoma)

CT (computed tomography)

MRI (magnetic resonance imaging)

T1CE (contrast-enhanced T1 weighted)

T1WI (T1 weighted imaging)

FLAIR(fluid attenuated inversion recovery )

DWI (diffusion weighted imaging)

ROC (Receiver operationg characteristic curve)

AUC(the area under the curve)

ACC(accuracy )

SEN (sensitivity)

SPE (specificity)

PPV (positive predictive palue)

NPV (negative predictive value)

Std(standard deviation)

CI(confidence interval)

TR (repetition time)

TE (echo time)

NSA (number of signal averages)

FOV (field of view)

GLCM (gray level co-occurrence matrix)

GLRLM (gray-level run-length matrix)

GLSZM (gray-level size zone matrix)

AIC (Akaike information criterion)

ROI (region of interest)

Acknowledgements

We would like to acknowledge the hard and dedicated work of all the staff that implemented the intervention and evaluation components of the study.

Author contributions

Conception and design of the research: LYS,XPY, JNW. Acquisition of data: LYS. Analysis and interpretation of the data: LYS, YZ,LHX,HM. Statistical analysis: QS,JLR.Writing of the manuscript:LYS. Critical revision of the manuscript for intellectual content:XPY, JNW. All authors read and approved the final draft.

Funding

The Outstanding Young Scientific Research and Innovation Team of Hebei University

(No. 605020521007);

The Hebei Provincial Government subsidizes the cultivation of outstanding talents in clinical medicine (No. 2021379).

Data availability

All data generated or analysed during this study are included in this article. Further enquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

This study was conducted with approval from the Ethics Committee of the Affiliated Hospital of Hebei University (No. HDFY-LL-2022-048). This study was conducted in accordance with the declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

LE RHUN E, GUCKENBERGER M, SMITS M, et al. EANO-ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up of patients with brain metastasis from solid tumours [J]. Ann Oncol, 2021, 32(11): 1332-47, https://doi.org/10.1016/j.annonc.2021.07.016.
FECCI P E, CHAMPION C D, HOJ J, et al. The Evolving Modern Management of Brain Metastasis [J]. Clin Cancer Res, 2019, 25(22): 6570-80, https://doi.org/10.1158/1078-0432.CCR-18-1624.
TESTA U, CASTELLI G, PELOSI E. Lung Cancers: Molecular Characterization, Clonal Heterogeneity and Evolution, and Cancer Stem Cells [J]. Cancers (Basel), 2018, 10(8), https://doi.org/10.3390/cancers10080248.
NICHOLSON A G, TSAO M S, BEASLEY M B, et al. The 2021 WHO Classification of Lung Tumors: Impact of Advances Since 2015 [J]. J Thorac Oncol, 2022, 17(3): 362-87, https://doi.org/10.1016/j.jtho.2021.11.003.
PARK H, SHOLL L M, HATABU H, et al. Imaging of Precision Therapy for Lung Cancer: Current State of the Art [J]. Radiology, 2019, 293(1): 15-29, https://doi.org/10.1148/radiol.2019190173.
KICKINGEREDER P, ANDRONESI O C. Radiomics, Metabolic, and Molecular MRI for Brain Tumors [J]. Semin Neurol, 2018, 38(1): 32-40, https://doi.org/10.1055/s-0037-1618600.
TOMASZEWSKI M R, GILLIES R J. The Biological Meaning of Radiomic Features [J]. Radiology, 2021, 299(2): E256, https://doi.org/10.1148/radiol.2021219005.
KNIEP H C, MADESTA F, SCHNEIDER T, et al. Radiomics of Brain MRI: Utility in Prediction of Metastatic Tumor Type [J]. Radiology, 2019, 290(2): 479-87, https://doi.org/10.1148/radiol.2018180946.
WANG G, WANG B, WANG Z, et al. Radiomics signature of brain metastasis: prediction of EGFR mutation status [J]. Eur Radiol, 2021, 31(7): 4538-47, https://doi.org/10.1007/s00330-020-07614-x.
CHEN B T, JIN T, YE N, et al. Radiomic prediction of mutation status based on MR imaging of lung cancer brain metastases [J]. Magn Reson Imaging, 2020, 69: 49-56, https://doi.org/10.1016/j.mri.2020.03.002.
ZHANG J, JIN J, AI Y, et al. Differentiating the pathological subtypes of primary lung cancer for patients with brain metastases based on radiomics features from brain CT images [J]. Eur Radiol, 2021, 31(2): 1022-8, https://doi.org/10.1007/s00330-020-07183-z.
LI Z, MAO Y, LI H, et al. Differentiating brain metastases from different pathological types of lung cancers using texture analysis of T1 postcontrast MR [J]. Magn Reson Med, 2016, 76(5): 1410-9, https://doi.org/10.1002/mrm.26029.
LIU J, CUI J, LIU F, et al. Multi-subtype classification model for non-small cell lung cancer based on radiomics: SLS model [J]. Med Phys, 2019, 46(7): 3091-100, https://doi.org/10.1002/mp.13551.
MARENTAKIS P, KARAISKOS P, KOULOULIAS V, et al. Lung cancer histology classification from CT images based on radiomics and deep learning models [J]. Med Biol Eng Comput, 2021, 59(1): 215-26, https://doi.org/10.1007/s11517-020-02302-w.
JIN K, HE K, TENG F, et al. Heterogeneity in primary tumors and corresponding metastases: could it provide us with any hints to personalize cancer therapy? [J]. Per Med, 2011, 8(2): 175-82, https://doi.org/10.2217/pme.10.81.
DREGELY I, PREZZI D, KELLY-MORLAND C, et al. Imaging biomarkers in oncology: Basics and application to MRI [J]. J Magn Reson Imaging, 2018, 48(1): 13-26, https://doi.org/10.1002/jmri.26058.
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, https://doi.org/10.1186/s12938-019-0744-0.
KHODABAKHSHI Z, MOSTAFAEI S, ARABI H, et al. Non-small cell lung carcinoma histopathological subtype phenotyping using high-dimensional multinomial multiclass CT radiomics signature [J]. Computers in Biology and Medicine, 2021, 136: 104752, https://doi.org/10.1016/j.compbiomed.2021.104752.
HAN Y, MA Y, WU Z, et al. Histologic subtype classification of non-small cell lung cancer using PET/CT images [J]. Eur J Nucl Med Mol Imaging, 2021, 48(2): 350-60, https://doi.org/10.1007/s00259-020-04771-5.
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, https://doi.org/10.21037/jtd-20-3358.
ZHU D, SHAO Y, YANG Z, et al. Magnetic resonance imaging characteristics of brain metastases in small cell lung cancer [J]. Cancer Medicine, 2023, 12(14): 15199-206, https://doi.org/10.1002/cam4.6206.
KAUFMANN T J, SMITS M, BOXERMAN J, et al. Consensus recommendations for a standardized brain tumor imaging protocol for clinical trials in brain metastases [J]. Neuro Oncol, 2020, 22(6): 757-72, https://doi.org/10.1093/neuonc/noaa030.
GROSSMAN R, HAIM O, ABRAMOV S, et al. Differentiating Small-Cell Lung Cancer From Non-Small-Cell Lung Cancer Brain Metastases Based on MRI Using Efficientnet and Transfer Learning Approach [J]. Technol Cancer Res Treat, 2021, 20: 15330338211004919, https://doi.org/10.1177/15330338211004919.
LI Y, YU R, CHANG H, et al. Identifying Pathological Subtypes of Brain Metastasis from Lung Cancer Using MRI-Based Deep Learning Approach: A Multicenter Study [J]. Journal of Imaging Informatics in Medicine, 2024, 37(3): 976-87, https://doi.org/10.1007/s10278-024-00988-0.
TULUM G. Novel radiomic features versus deep learning: differentiating brain metastases from pathological lung cancer types in small datasets [J]. BRITISH JOURNAL OF RADIOLOGY, 2023, 96(1146), https://doi.org/10.1259/bjr.20220841.
JIANG Y, ZHANG Z, WANG W, et al. Biology-guided deep learning predicts prognosis and cancer immunotherapy response [J]. Nat Commun, 2023, 14(1): 5135, https://doi.org/10.1038/s41467-023-40890-x.
BEKAERT L, EMERY E, LEVALLET G, et al. Histopathologic diagnosis of brain metastases: current trends in management and future considerations [J]. Brain Tumor Pathol, 2017, 34(1): 8-19, https://doi.org/10.1007/s10014-016-0275-3.
DERKS S H A E, VELDT A A M V D, SMITS M. Brain metastases: the role of clinical imaging [J]. The British Journal of Radiology, 2022, 95(1130): 20210944, https://doi.org/10.1259/bjr.20210944.
DEIKE-HOFMANN K, THüNEMANN D, BRECKWOLDT M O, et al. Sensitivity of different MRI sequences in the early detection of melanoma brain metastases [J]. PLoS One, 2018, 13(3): e0193946, https://doi.org/10.1371/journal.pone.0193946.
AHN S J, KWON H, YANG J J, et al. Contrast-enhanced T1-weighted image radiomics of brain metastases may predict EGFR mutation status in primary lung cancer [J]. Sci Rep, 2020, 10(1): 8905, https://doi.org/10.1038/s41598-020-65470-7.
YOUNG J R, RESSLER J A, MORTIMER J E, et al. Association of lesion contour and lesion composition on MR with HER2 status in breast cancer brain metastases [J]. Magnetic Resonance Imaging, 2023, 96: 60-6, https://doi.org/10.1016/j.mri.2022.11.009.
XIANG X, LI X, LIN H, et al. Amide proton transfer-weighted MRI in predicting pathological types of brain metastases in lung Cancer [J]. Magn Reson Imaging, 2024, 108: 59-66,https://doi.org/10.1016/j.mri.2024.01.014.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Identification of the Pathological Types of Brain Metastasis from Lung Cancer Based on Multiparametric MRI Radiomics: A Feasibility Study

Status:

Version 1

Abstract

Objectives

Methods

Results

Conclusion

Figures

Introduction

Materials and methods

Subjects

MR images acquisition

Tumor segmentation and features extraction

Radiomics features selection and model building

Statistical analysis

Results

Patients’ characteristics

Classification of performance of the models

Performance of models to differentiate NSCLC from SCLC

Performance of models to differentiate AD from NAD

Discussion

Conclusion

Abbreviations

Declarations

Ethics approval and consent to participate

References

Additional Declarations

Status:

Version 1