Radiomics of voxel-wise DCE-MRI time-intensity-curve profile map enable quantitative assessment of temporal and spatial hemodynamic heterogeneity within breast lesions

doi:10.21203/rs.3.rs-4020070/v1

Download PDF

Research Article

Radiomics of voxel-wise DCE-MRI time-intensity-curve profile map enable quantitative assessment of temporal and spatial hemodynamic heterogeneity within breast lesions

https://doi.org/10.21203/rs.3.rs-4020070/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

To investigate the usefulness of radiomics analysis based on voxel-wise mapping of DCE-MRI time-intensity-curve (TIC) profiles in quantifying temporal and spatial hemodynamic heterogeneity.

Methods

From December 2018 to August 2022, 428 patients with 639 breast lesions were retrospectively enrolled. The TIC profile of each voxel within the manually segmented 3D lesion was categorized into 19 subtypes based on wash-in rate (nonenhanced, slow, medium, and fast), wash-out enhancement (persistent, plateau, and decline), and wash-out stability (steady and unsteady). Three feature sets were calculated separately, including composition ratio (type-19) and radiomics features (type-19-radiomics) of 19 TIC profile subtypes, and radiomics features based on third-phase DCE-MRI images (phase-3-radiomics). Using support vector machine, four models (type-19, type-19-radiomics, type-19-combined, and phase-3-radiomics) were constructed to distinguish benign and malignant breast lesions.

Results

In differentiating benign and malignant lesions, both cross-validation and independent testing showed that type-19-combined model significantly outperformed phase-3-radiomics model (AUC = 0.906 vs. 0.823, P < 0.001, AUC = 0.867 vs. 0.762, P = 0.026). However, in cross-validation and testing, no significant difference in performance was observed between phase-3-radiomics model and type-19 model (P = 0.577 and 0.085), between phase-3-radiomics model and type-19-radiomics model (P = 0.182 and 0.200), or between type-19-radiomics model and type-19 model (P = 0.073 and 0.454).

Conclusions

In addition to radiomics analysis based on a single phase DCE-MRI, radiomics analysis of voxel-wise DCE-MRI time-intensity-curve (TIC) profiles map enables quantifying temporal and spatial hemodynamic heterogeneity simultaneously, thereby aiding in the differentiation of breast lesions.

dynamic contrast-enhanced MRI (DCE-MRI)

time-intensity-curve (TIC)

Breast cancer

Radiomics

Hemodynamic Heterogeneity

• Radiomics analysis based on a voxel-wise TIC profile map provides a novel model-free data-driven approach to simultaneously quantify spatial and temporal heterogeneity within breast lesions.

• This approach might hold the potential to characterize and differentiate breast lesions.

Breast cancer has become the most prevalent cancer globally with high mortality.¹ In 2020, 2.3 million women were diagnosed with breast cancer and there were 685000 deaths worldwide.² Breast cancer exhibits tumor heterogeneity in angiogenesis at the inter-tumoral and intra-tumoral levels, which plays a key role in biological behavior, tumor growth and metastasis, treatment response, and prognosis.³ However, histology based on a portion of the lesion specimen could not show the hemodynamic heterogeneity of the entire lesion and also could not characterize the hemodynamics of the lesion in a physiological way. Therefore, a non-invasive tool that could capture, visualize, and quantify 3D intra-tumoral hemodynamic heterogeneity is needed to characterize the whole lesion in a physiological way.

For its superb soft tissue contrast, MRI is routinely used for breast lesion detection, characterization, treatment response evaluation, prognosis prediction, etc.⁴ Among the breast MRI protocols, DCE-MRI has become the most important sequence for its capacity to reveal hemodynamic heterogeneity by capturing the dynamic signal changes over time after injecting a contrast agent.⁵ However, currently available methods (qualitative, semi-quantitative, and model-based quantitative analysis) have limitations in the clinical setting. Qualitative analysis lacks comprehensive information on intra-tumoral spatial hemodynamic heterogeneity, suffers from non-whole-tumor coverage, region of interest (ROI) selection bias, and inter-observer variability, and fails to reflect spatial heterogeneity quantitatively.⁶ Similarly, semi-quantitative analysis is constrained by ROI selection bias, average effects, and inconsistent definitions across studies.⁷ Moreover, for quantitative analysis, the complexity and inconsistency of various quantitative ideal models, such as ETK, susceptible to multiple factors (i.e., MRI parameters, pre-contrast tissue T1 value, arterial input function, noise level, image deformation, and/or low temporal resolution, etc.) and artifacts, hinder their clinical applicability.^{8, 9}

To bypass the complexities associated with the existing model, we previously proposed a novel model-free data-driven approach to generate voxel-wise mapping of diverse DCE-MRI TIC profiles, which facilitates the intuitive visualization of intra-lesion hemodynamic heterogeneity. Furthermore, we introduced the composition ratio of these different voxel-wise TIC profiles within a complete 3D lesion as a novel feature set, which enables quantifying temporal hemodynamic heterogeneity, and then validated its value for differentiating malignant and benign breast lesions.¹⁰ However, notably, the composition ratio could not reflect the spatial distribution heterogeneity of these diverse TIC profiles.

Hence, in this study, we intend to extract radiomics features from voxel-wise TIC profile maps to quantify the spatial distribution heterogeneity of various TIC profile subtypes. Theoretically, these TIC profile subtypes-based radiomics features could simultaneously reflect temporal and spatial hemodynamic heterogeneity. In addition, the value of these voxel-wise TIC profile subtypes map-based radiomics features were explored in distinguishing benign and malignant breast lesions.

The current research study received approval from the local institutional review board (IRB) with a waiver of informed consent for all recruited participants. Spanning from December 2018 to August 2022, the study retrospectively enrolled 428 female patients harboring 639 breast lesions, all of whom had previously undergone baseline DCE-MRI. The inclusion criteria encompassed the following: (1) verification of the histology of each breast lesion via either biopsy or surgical specimen, or benignity verified through at least 1-year follow-up of each benign breast lesion; (2) DCE-MRI performed within a maximum of 2 weeks prior to treatment; (3) absence of biopsy or neoadjuvant chemotherapy prior to DCE-MRI, and (4) acquisition of a minimum of 9 phases in the baseline DCE-MRI scan. The quality of the acquired DCE-MRI scans underwent stringent evaluation by breast imaging specialists, with those exhibiting degraded quality, such as severe noise, motion artifacts, or geometric distortion, being excluded from the subsequent analysis. Figure 1 provides an overview of the patient selection process.

MRI protocols

The research employed a 3-Tesla MR scanner in combination with an 8-channel phase-array breast coil. The DCE-MRI protocol employed a three-dimensional T1-weighted fast spoiled gradient echo sequence. These specific imaging parameters included a 4.5 ms repetition time (TR), an 2.1 ms echo time (TE), a field of view measuring 360 × 360 mm, an 320 × 320 image matrix, a 1.4 mm slice thickness with no inter-slice gap, and a 12° flip angle. The DCE-MRI scan consisted of a minimum of 9 phases, with each phase having a duration of 35–55 seconds. These phases comprised one pre-contrast and at least eight post-contrast phases. During the DCE-MRI scan, Gadoteric acid meglumine, was administered intravenously at a rate of 2 mL/s with a total dose equivalent to 0.1 mmol/kg of body weight. Subsequently, a 20 mL saline flush was administered.

Image preprocessing

Each breast lesion underwent manual segmentation on the third phase of DCE-MR images slice-by-slice by two junior radiologists, M.W. and Y.R., who had more than five years of experience in breast MRI. After this initial process, two senior radiologists (J.W. and Z.L.) with more than ten years of expertise in breast MRI reviewed the segmented lesions. Any discrepancy in segmentation was resolved through discussion. The pre-contrast image (S0) was utilized to normalize the subsequent corresponding post-contrast images (S) voxel-by-voxel. Subsequently, TIC profile for each voxel was generated for each segmented 3D lesion voxel-by-voxel using an in-house Python platform (see Fig. 2).

Definition of type‑19

The TIC profiles were categorized into 19 unique subtypes (3 × 3 × 2 + 1) based on the attributes of their wash-in rate (non-enhanced, slow, medium, and fast), wash-out enhancement (persistent, plateau, and decline), and wash-out stability (steady and non-steady). This categorization covers all possible types of TIC profile morphologies to ensure inclusivity (see Fig. 3). A detailed definition of type-19 can be found in our previous work.¹⁰

Feature extraction

1) type-19 feature set: The TIC profile of each voxel within the 3D lesion was classified into one subtype of 19 defined TIC profiles voxel by voxel, and the percentage of voxels with the same TIC profile accounting for all the voxels (composition ratio) represented a feature set (type-19).

2) phase-3-radiomics feature set and type-19-radiomics feature set: Firstly, a new 3D matrix for each segmented lesion was generated by assigning the voxel with the corresponding number of type-19 profiles. Then, radiomics features based on manually segmented 3D lesions on original phase-3 DCE-MR images (phase-3-radiomics) and newly generated TIC-profile number matrix (type-19-radiomics) were extracted using Pyradiomics package¹¹ in an in-house python platform, respectively. We obtained 1130 radiomics features for phase-3-radiomics set and type-19-radiomic feature set, respectively, including original features (n = 107), log features (n = 279), and wavelet features (n = 744). The original features are further categorized into shape features (n = 14), first-order features (n = 18), and texture features (n = 75). Log features are obtained by calculating sigma values (1.0mm, 3.0mm, 5mm) from first-order and texture features ((18 + 75) × 3 = 279). Wavelet features are obtained from first-order and texture features by combining high-pass and low-pass filters separately in each of the three dimensions (93 × 8 = 744).

Feature selection and ranking

With a hold-out scheme (7: 3), the dataset was split into a training set and a testing set. Initially, a two-independent sample student-t-test was conducted to select features that exhibited statistically significant differences between malignant and benign lesions (P < 0.05). Then, the least absolute shrinkage and selection operator (Lasso) regression algorithm¹² was utilized to eliminate irrelevant features, prevent overfitting, and enhance interpretability. The Lasso algorithm determined the most informative features based on the lambda (λ) value (a tuning parameter) with minimal deviance. The impact of each feature on the classification performance was assessed using its Lasso coefficient, an independent filter model designed to avoid interference with the classification learning algorithm.¹³

Clinical validation and model building

The clinical relevance of the type-19-radiomics feature set was validated to differentiate malignant and benign lesions. Four models were built based on selected features, including phase-3-radiomics model, type-19 model, type-19-radiomics model, and type-19-combined model (type-19 features + selected type-19-radiomics features). Support vector machine (SVM) was used to train the machine learning model based on different feature sets. 10-fold cross-validation scheme was used and the training set was firstly split into 10 folds to make sure the model performance can be compared across different models. Receiver operating characteristic (ROC) curve analysis was used to evaluate the performance on the 10-fold cross-validation set and independent testing set with the area under the curve (AUC) compared using the DeLong test.¹⁴ Two-tailed P < 0.05 indicated statistical significance.

A total of 428 female patients with 639 breast lesions who had undergone baseline DCE-MRI were finally selected for data analysis. No patients enrolled had both benign and malignant lesions. The training set contained 97 benign lesions and 351 malignant lesions. The testing set had 47 benign lesions and 145 malignant lesions. Patient clinicopathologic characteristics are summarized in Table 1.

Table 1

Clinical-pathologic characteristics of enrolled patients
	Malignant cases (n = 288)	Benign cases (n = 140)
Age (years)	49.5 ± 11.41	42.53 ± 11.19
Number of lesions	495	144
Maximal diameter (cm)	2.34 ± 1.67	1.41 ± 1.11
Histology	Invasive breast carcinoma of non-special type (n = 442)	Adenosis (n = 14)
	In-situ carcinoma (n = 28)	Fibroadenoma (n = 52)
	Papillary carcinoma (n = 6)	Inflammatory lesion (n = 18)
	Mucinous carcinoma (n = 10)	Intraductal papilloma (n = 17)
	Tubular carcinoma (n = 1)	Adenomyoepitheliomatous change (n = 2)
	Invasive lobular carcinoma (n = 7)	Mammary hamartomas (n = 1)
	medullary carcinoma (n = 1)	Cyst (n = 3)
		Borderline lobe tumor (n = 1)
		Not available (n = 36)
Ki-67 status
Low (< 20%)	118	--
High (≥ 20%)	375	--
Not available	2	--
Histologic grade
I	26	--
II	223	--
III	148	--
Not available	99	--
Molecular subtype
Luminal A	35	--
Luminal B	315	--
HER2-enriched	111	--
Triple negative	33	--
Not available	1

Type-19 feature set

As shown in Fig. 4, benign lesions exhibited a significantly higher percentage of the 1st, 2nd, and 19th TIC subtypes and a lower percentage of the 9th, 10th, 11th, 12th, 15th, 16th, 17th, and 18th TIC subtypes compared to malignant lesions (all P < 0.05). In benign breast lesions, the predominant TIC profiles consisted of slow wash-in and persistent enhancement TIC profiles (2nd + 1st subtypes), followed by the medium wash-in and persistent enhancement with oscillation (8th subtype) and no-enhancement TIC profiles (19th subtype). In contrast to benign lesions with only a few predominant TIC profile subtypes, malignant lesions showed a more evenly distribution of TIC profiles.

Radiomics Feature selection and ranking

For type-19-radiomics features, 767 features were selected after t-test. Then, 15 features (12 texture features and three first-order features) were selected after Lasso regression (|coefficient| > 0.02). For phase-3-radiomics features, 823 and 17 radiomics features (5 first-order features, two shape features, and 10 texture features) were sequentially selected after t-test and Lasso regression (|coefficient| > 0.02). Regarding feature importance ranking, the top 5 features with the highest absolute coefficient were all texture features with log transformation (n = 3) and wavelet transformation (n = 2) for type-19-radiomics feature set. In contrast, the top 5 features with the highest absolute coefficient included four first-order features (wavelet transformation [n = 3] and log transformation [n = 1]) and one texture feature with log transformation for phase-3-radiomics feature set (Fig. 5).

Model performance

In differentiating the lesion malignancy, both 10-fold cross-validation (Table 2) and independent testing (Table 3) showed that the type-19-combined model significantly outperformed the phase-3-radiomics model (AUC = 0.906 vs. 0.823, P < 0.001, AUC = 0.867 vs. 0.762, P = 0.026). However, in 10-fold cross-validation (Table 2) and independent testing (Table 3), there was no significant difference in performance observed between the phase-3-radiomics model and the type-19 model (AUC = 0.823 vs. 0.809, P = 0.577, AUC = 0.762 vs. 0.844, P = 0.085), between the phase-3-radiomics model and the type-19-radiomics model (AUC = 0.823 vs. 0.854, P = 0.182, AUC = 0.762 vs. 0.816, P = 0.200), or between the type-19-radiomics model and the type-19 model (AUC = 0.854 vs. 0.809, P = 0.073, AUC = 0.816 vs. 0.844, P = 0.454) (Fig. 6).

Table 2

Model performance in 10-fold cross-validation
SVM	Accuracy	Sensitivity	Specificity	Precision	F1-score	AUC
phase-3-radiomics	0.828	0.959	0.350	0.842	0.896	0.823
type-19-radiomics	0.866	0.969	0.499	0.874	0.918	0.854
type-19	0.864	0.958	0.528	0.879	0.916	0.809
type-19-combined	0.882	0.966	0.579	0.892	0.927	0.906

Table 3

The model performance in independent testing
SVM	Accuracy	Sensitivity	Specificity	Precision	F1-score	AUC
phase-3-radiomics	0.802	0.683	0.723	0.796	0.883	0.761
type-19-radiomics	0.823	0.821	0.723	0.828	0.892	0.816
type-19	0.813	0.745	0.830	0.843	0.882	0.844
type-19-combined	0.854	0.731	0.851	0.863	0.909	0.868

In this study, we investigated the usefulness of radiomics features based on the TIC profile map in simultaneously reflecting temporal and spatial hemodynamic heterogeneity and further validated its potential application in differentiating malignant and benign breast lesions. The type-19-combined (type-19 + type-19-radiomics) model significantly outperformed the phase-3-radiomics model in distinguishing malignant breast lesions from benign ones, suggesting quantification of spatial distribution heterogeneity of TIC profiles using radiomics analysis provides a novel approach to quantify temporal and spatial hemodynamic heterogeneity within breast lesions simultaneously.

Previously, most studies have extracted the radiomics features from a single phase of DCE-MRI and demonstrated discriminative power in differentiating malignant and benign lesions.^15–17 For instance, Zhang et al. extracted radiomics features from the 8th phase of DCE-MRI images, demonstrating good discriminatory ability to differentiate benign and malignant breast lesions.¹⁵ Militello et al. built a radiomics predictive model based on the phase with the most marked enhancement of breast DCE-MRI, which showed promising results in accuracy and specificity in differentiating malignant from benign breast lesions.¹⁶ In line with these studies, our study also showed good performance of radiomics features extracted from the 3rd phase of DCE-MRI in discriminating malignant and benign lesions. However, although radiomics features show good performance by exploiting spatial hemodynamic heterogeneity, this approach uses a single phase of post-contrast DCE-MRI. This approach could not take advantage of the temporal hemodynamic heterogeneity of the multiple phases of DCE-MRI simultaneously.

Previously, we proposed a data-driven and model-free approach to classify voxel-wise TIC profiles into 19 TIC profile subtypes. We generated the color-coded TIC profile subtype distribution map (type-19 map) to visualize the intra-lesional hemodynamic heterogeneity intuitively. In addition, we proposed and validated the composition ratio of TIC profile subtypes as a new feature set to differentiate malignant and benign lesions. However, the concept of composition ratio could only reflect temporal hemodynamic heterogeneity, but it could not extract the spatial distribution heterogeneity of diverse TIC profile subtypes. Radiomics analysis is a perfect way to quantify the spatial heterogeneity of gray-scale images.¹⁸ That is why, in this study, we incorporated radiomics analysis to extract the spatial distribution heterogeneity of TIC profile subtypes to quantify temporal and spatial hemodynamic heterogeneity simultaneously. In our study, both type-19 and type-19 radiomics features performed well in distinguishing malignant and benign lesions, demonstrating another effective way to quantify temporal and spatial hemodynamic heterogeneity.

In our study, texture features were predominant in the phase-3-radiomics feature set, suggesting spatial heterogeneity makes the dominant contribution to the differentiation, which is in line with previous studies.^{17, 19} For instance, a study by Roberta et al. showed 32 out of 37 robust radiomics features were texture features which contributed to good performance in the discrimination of benign and malignant lesions.¹⁷ In terms of TIC profile subtypes, it has been proven that benign lesions mostly exhibit TIC with slow wash-in and persistent enhancement. In contrast, malignant lesions tend to show fast wash-in and decline wash-out TIC with more variability.^{20, 21} Consistent with our previous study,¹⁰ there was a significantly higher percentage of the 1st and 2nd subtypes for benign lesions and the 11th, 12th, 17th and 18th subtypes for malignant lesions. For the type-19-radiomics feature set, most of the selected features were also texture features, suggesting spatial TIC profile distribution heterogeneity is vital in malignancy differentiation, which needs further confirmation in future studies.

The present study possesses several limitations. Firstly, this study was conducted retrospectively and solely within a singular institution, suggesting the necessity for further external validation of the proposed model. Secondly, the ROIs were manually delineated on a per-slice basis, which can be time-consuming and subject to inter-observer variability. Consequently, adopting a more reliable and objective automated segmentation technique holds the potential to be a better solution in the future. Lastly, it should be acknowledged that a direct comparison between the model performance based on the type-19-radiomics features with the quantitative analysis-based model and semi-quantitative analysis-based model was not conducted due to disparities in MRI scan parameters and the absence of pre-contrast T1 values of the tissue and arterial input function.

In conclusion, radiomics analysis based on a voxel-wise intra-lesion color-coded TIC profile map provides a novel model-free data-driven approach to quantify spatial and temporal heterogeneity within breast lesions.

Ethics approval and consent to participate

Institutional Review Board approval was obtained. Written informed consent was waived by the Institutional Review Board.

Competing interests

The authors declare no competing interests

Funding

This work was supported by the National Key Technology Research and Development Program of China (2023YFC3402802 and 2021YFF0501502), Natural Science Foundation of Guangdong Province-Outstanding Youth Project (2023B1515020002), National Natural Science Foundation of China (No. 82302153), Cancer Hospital Chinese Academy of Medical Sciences-Shenzhen Hospital Cooperative Fund (CFA202202022), Shenzhen Clinical Research Center for Cancer (No. [2021] 287) and Shenzhen High-level Hospital Construction, Key Laboratory for Magnetic Resonance and Multimodality Imaging of Guangdong Province (2023B1212060052), and Central guidance for local science and technology development project (ZYYD2023D02).

Acknowledgements

We would like to thank Dr. Long Qian and Dr. Wei Cui for their technical support.

Author Contribution

Zhou Liu and Meng Wang wrote the main manuscript text, Meng Sun and Bingyu Yao conducted experiments and drew Figures 1-6，Ya Ren and Wei Cui prepared table 1，Qian Yang prepared table 2，Long Qian prepared table 3. All authors reviewed the manuscript.

Giaquinto AN, Sung H, Miller KD, Kramer JL, Newman LA, Minihan A, et al. Breast Cancer Stat 2022 CA Cancer J Clin. 2022;72(6):524–41.
Coles CE, Anderson BO, Cameron D, Cardoso F, Horton R, Knaul FM, et al. The Lancet Breast Cancer Commission: tackling a global health, gender, and equity challenge. Lancet. 2022;399(10330):1101–3.
Lee JY, Lee KS, Seo BK, Cho KR, Woo OH, Song SE, et al. Radiomic machine learning for predicting prognostic biomarkers and molecular subtypes of breast cancer using tumor heterogeneity and angiogenesis properties on MRI. Eur Radiol. 2022;32(1):650–60.
Mann RM, Cho N, Moy L, Breast MRI. State of the Art. Radiology. 2019;292(3):520–36.
Leithner D, Wengert GJ, Helbich TH, Thakur S, Ochoa-Albiztegui RE, Morris EA, et al. Clinical role of breast MRI now and going forward. Clin Radiol. 2018;73(8):700–14.
Goto M, Ito H, Akazawa K, Kubota T, Kizu O, Yamada K, et al. Diagnosis of breast tumors by contrast-enhanced MR imaging: comparison between the diagnostic performance of dynamic enhancement patterns and morphologic features. J Magn Reson Imaging. 2007;25(1):104–12.
Liu HL, Zong M, Wei H, Lou JJ, Wang SQ, Zou QG, et al. Differentiation between malignant and benign breast masses: combination of semi-quantitative analysis on DCE-MRI and histogram analysis of ADC maps. Clin Radiol. 2018;73(5):460–6.
Winfield JM, Payne GS, Weller A, deSouza NM, DCE-MRI. DW-MRI, and MRS in Cancer: Challenges and Advantages of Implementing Qualitative and Quantitative Multi-parametric Imaging in the Clinic. Top Magn Reson Imaging. 2016;25(5):245–54.
Garcia-Figueiras R, Baleato-Gonzalez S, Padhani AR, Luna-Alcala A, Vallejo-Casas JA, Sala E, et al. How clinical imaging can assess cancer biology. Insights Imaging. 2019;10(1):28.
Liu Z, Yao B, Wen J, Wang M, Ren Y, Chen Y et al. Voxel-wise mapping of DCE-MRI time-intensity-curve profiles enables visualizing and quantifying hemodynamic heterogeneity in breast lesions. Eur Radiol. 2023 Aug 11.
van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 2017;77(21):e104–7.
Liu Y, Xu P, Cao C, Shan B, Zhu K, Ma Q et al. A comparative evaluation of machine learning algorithms and an improved optimal model for landslide susceptibility: a case study. Geomatics, Natural Hazards and Risk. 2021;12(1):1973–2001.
Meng L, Zheng T, Wang Y, Li Z, Xiao Q, He J, et al. Development of a prediction model based on LASSO regression to evaluate the risk of non-sentinel lymph node metastasis in Chinese breast cancer patients with 1–2 positive sentinel lymph nodes. Sci Rep. 2021;11(1):19972.
Guo L, Du S, Gao S, Zhao R, Huang G, Jin F et al. Delta-Radiomics Based on Dynamic Contrast-Enhanced MRI Predicts Pathologic Complete Response in Breast Cancer Patients Treated with Neoadjuvant Chemotherapy. Cancers (Basel). 2022;14(14).
Zhang Q, Peng Y, Liu W, Bai J, Zheng J, Yang X, et al. Radiomics Based on Multimodal MRI for the Differential Diagnosis of Benign and Malignant Breast Lesions. J Magn Reson Imaging. 2020;52(2):596–607.
Militello C, Rundo L, Dimarco M, Orlando A, Woitek R, D'Angelo I, et al. 3D DCE-MRI Radiomic Analysis for Malignant Lesion Prediction in Breast Cancer Patients. Acad Radiol. 2022;29(6):830–40.
Fusco R, Piccirillo A, Sansone M, Granata V, Vallone P, Barretta ML et al. Radiomic and Artificial Intelligence Analysis with Textural Metrics, Morphological and Dynamic Perfusion Features Extracted by Dynamic Contrast-Enhanced Magnetic Resonance Imaging in the Classification of Breast Lesions. Appl Sci. 2021;11(4).
Satake H, Ishigaki S, Ito R, Naganawa S. Radiomics in breast MRI: current progress toward clinical application in the era of artificial intelligence. Radiol Med. 2022;127(1):39–56.
Wang TC, Huang YH, Huang CS, Chen JH, Huang GY, Chang YC, et al. Computer-aided diagnosis of breast DCE-MRI using pharmacokinetic model and 3-D morphology analysis. Magn Reson Imaging. 2014;32(3):197–205.
El Khouli RH, Macura KJ, Kamel IR, Jacobs MA, Bluemke DA. 3-T dynamic contrast-enhanced MRI of the breast: pharmacokinetic parameters versus conventional kinetic curve analysis. AJR Am J Roentgenol. 2011;197(6):1498–505.
Cheng Z, Wu Z, Shi G, Yi Z, Xie M, Zeng W, et al. Discrimination between benign and malignant breast lesions using volumetric quantitative dynamic contrast-enhanced MR imaging. Eur Radiol. 2018;28(3):982–91.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
02 Nov, 2024
Reviews received at journal
24 Sep, 2024
Reviewers agreed at journal
25 Aug, 2024
Reviewers invited by journal
07 Apr, 2024
Editor assigned by journal
13 Mar, 2024
Submission checks completed at journal
13 Mar, 2024
First submitted to journal
06 Mar, 2024

You are reading this latest preprint version

Radiomics of voxel-wise DCE-MRI time-intensity-curve profile map enable quantitative assessment of temporal and spatial hemodynamic heterogeneity within breast lesions

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Clinical Relevance Statement

Introduction

Materials and Methods

MRI protocols

Image preprocessing

Definition of type‑19

Feature selection and ranking

Clinical validation and model building

Result

Type-19 feature set

Radiomics Feature selection and ranking

Model performance

Discussion

Declarations

References

Additional Declarations

Status:

Version 1