Improved diagnostic value of whole-lesion histogram and texture analyses on multiparametric breast MRI for papillary neoplasms with non-mass enhancement

Introduction

As defined in the 2019 World Health Organization (WHO) classification system for breast tumors, papillary breast neoplasms comprise a range of pathological entities, spanning from benign papilloma to invasive papillary carcinoma (1). Distinguishing between benign and malignant entities is a critical step in the diagnostic evaluation of breast papillary neoplasms (2). The management of these lesions is guided by their histological diagnosis (1,3,4). Currently, the diagnosis of breast papillary neoplasms primarily relies on core needle biopsy or postoperative pathological examination. However, biopsy procedures are invasive and may not fully capture the tumor’s heterogeneity due to limited tissue sampling and overlapping pathologic findings, especially for diffuse non-mass enhancement (NME) lesions (2). The wide spectrum of pathology associated with papillary tumors results in a multitude of imaging manifestations, posing a challenge for diagnosis. This challenge is particularly pronounced in the case of papillary neoplasms on magnetic resonance imaging (MRI) with NME. NME is more diffuse and can mimic benign processes like fibrocystic changes, inflammation, or even normal hormonal variations (5). The diffuse and subtle nature of the NME lesion can obscure key histological features, making it harder to differentiate between these entities (5). So, it remains a complex aspect of modern imaging diagnosis to distinguishing between benign and malignant areas of enhancement. Thus, we aim to identify a non-invasive MRI-based method for differentiating benign and malignant papillary lesions presenting as NMEs.

Currently, there has been increasing interest in histogram and texture analysis, which can provide a more comprehensive understanding of tumor heterogeneity. Histogram analysis allows for the estimation of the probability distribution of a continuous variable. Past research has shown that this technique is effective in classifying breast tumors based on diffusion-weighted MRI (6-8). Texture analysis, on the other hand, by evaluating the distribution of gray levels, coarseness, and regularity in a tumor, can offer a measurement of tumor heterogeneity using the diagnostic images obtained during routine clinical practice, without requiring extra imaging or invasive procedures. Breast MRI texture analysis has been utilized to aid in the assessment of survival outcomes (9), distinguish between benign and malignant lesions (10), forecast treatment response (11,12), and differentiate among various subtypes of breast cancer (13).

As far as we know, there has not been any published study that has explored the differential diagnosis of breast papillary neoplasms by utilizing histogram and texture analysis. Earlier research has revealed that employing a whole-tumor histogram and texture analysis software for multiparametric magnetic resonance (MR) images, such as diffusion-weighted imaging (DWI) and CAIPIRINHA-Dixon-TWIST-volume-interpolated breath-hold examination (VIBE) (CDTV), can be effective in distinguishing triple-negative breast cancer (14) and differentiating breast lesions and breast cancer molecular subtypes (15). Moreover, the most recent research also indicates that it can be particularly useful in differentiating suspicious breast tumors that are smaller than 1 cm (16).

In this study, we used the same analysis software to analyze DWI sequences, apparent diffusion coefficient (ADC) map, and subtraction (SUB) MR images. This study aims to assess the diagnostic efficacy of quantitative parameters derived from histogram and texture analysis in comparison to qualitative parameters obtained from MRI, aiming to distinguish between benign and malignant papillary neoplasms on MRI with NME, in order to avoid overtreatment. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-128/rc).

Methods

Patients

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of Ruijin Hospital (Shanghai Jiao Tong University School of Medicine) [No. 2023(333)]. Individual consent for this retrospective analysis was waived due to the retrospective nature. Between December 2016 and January 2023, we conducted breast MRI examinations on 381 consecutive women with surgically confirmed papillary lesions, all the MRI examinations were performed before biopsies. We excluded patients who met the following criteria: no preoperative breast MRI within 2 weeks before surgery (n=32); lesions poorly identifiable on MRI (n=17); prior history of breast surgery or chemoradiotherapy (n=58); significant motion artifacts on imaging (n=6). Then papillary lesions with focus or masses (n=170) were excluded. In the end, a total of 98 women with 98 lesions were enrolled in the study.

MR scanning

We performed the study on a 1.5-T MAGNETOM Aera MRI scanner (Siemens Healthcare, Berlin, Germany) equipped with an 18-channel phased-array dedicated breast coil. Our standard acquisition protocol for breast MRI included fat-suppressed T2-weighted fast spin-echo images, DWI images, ADC map, and dynamic contrast-enhanced (DCE) images.

DWI was performed prior to contrast agent administration utilizing a readout-segmented echo-planar imaging (EPI) sequence with the following parameters: repetition time (TR)/echo time (TE) =3,800/62 ms; field of view (FOV) =36 cm; matrix =384×384; slice thickness =5 mm; number of slices =20; b value =50 and 800 s/mm². The ADC map was automatically calculated using a mono-exponential fitting algorithm incorporating all b-values, performed by the scanner’s integrated Syngo software.

DCE images were acquired using fat-suppressed T1-weighted sequences on five consecutive occasions before and after contrast injection, with the following parameters: TR/TE =4.48/1.85 ms; FOV =36 cm; matrix =384×384; slice thickness =1.5 mm; number of slices =104. SUB MR images were obtained by digitally subtracting the unenhanced T1-weighted sequence from the initial phase-enhanced T1-weighted sequence performed after gadolinium administration.

Clinical and imaging characteristics analysis

Cases underwent random assignment and thorough evaluation, aligning with the 2013 edition of the Breast Imaging Reporting and Data System (BI-RADS) lexicon (17). Two breast radiologists (Q.F. and K.S.), one with 3 years and the other with a decade of specialized experience in breast MRI, were kept unaware of the pathological results to uphold an impartial assessment. They have read images independently. In instances of discordance in their interpretations, they engaged in collaborative discussions to reach a consensus.

Feature extraction and selection

Histogram and texture analyses were performed utilizing the prototype MR Multiparametric Analysis software (Siemens Healthineers). The above two radiologists performed the analyses in four steps:

• Data loading: we loaded DWI images with b-values of 50/800 s/mm², ADC map, and SUB MR onto the software.
• Seed point annotation: our radiologists manually placed foreground and background seed markers within and outside the breast tumor across all three multiplanar reconstruction (MPR) planes of the provided images.
• Segmentation: the tumor boundaries were automatically delineated through a random walker approach initiated from the predefined seed points, with subsequent manual refinement performed when necessary to ensure segmentation accuracy. The software then automatically propagated the final 3D-segmented volumes created on the post-contrast images to the DWI_b=50, DWI_b=800, ADC map, and SUB MR.
• Histogram and texture analyses: our software automatically performed whole-tumor histogram and texture analyses on the aforementioned images with the push of a button. We quantitatively extracted eleven distinct image features, comprising seven histogram-derived statistical parameters and four textural characteristics. The histogram-based metrics included: mean intensity value, standard deviation (SD), median value, 5^th and 95^th percentiles, skewness coefficient, and kurtosis measure. The textural features encompassed: image entropy, contrast index, differential entropy (diff-entropy), and differential variance (diff-variance).

The steps of II, III, and IV were illustrated in Figure 1.

Figure 1 Workflow of the histogram and texture analysis. (A-C) Manually drew foreground and background seed points inside (marked in green) and outside (marked in red) the three multiplane reconstruction planes of post-contrast images. (D-G) 3D segmentations generated based on the post-contrast images. Histogram was generated. 3D, three-dimensional.

The measurements obtained from the two radiologists were averaged for subsequent statistical analysis. Intra-observer agreement was evaluated by having Q.F. repeat the measurements after a 2-week interval, blinded to the initial results. Inter-observer variability was assessed by comparing the baseline measurements from Q.F. with those obtained independently by K.S.

After extracting 44 features, least absolute shrinkage and selection operator (LASSO) was applied to optimize feature selection, as depicted in Figure 2.

Figure 2 LASSO was used to select features. Seventeen features corresponded to the minimum error. LASSO, least absolute shrinkage and selection operator.

The quantitative model’s reliability was assessed through 10-fold cross-validation, where each round utilized 90% of the dataset for training purposes and the remaining 10% for evaluation.

Statistical analysis

Continuous data were presented as mean ± SD, whereas categorical data were summarized using counts and proportions. For univariate comparisons, normally distributed continuous variables were assessed with Student’s t-tests, while non-normally distributed variables were evaluated using the Mann-Whitney U test. Categorical variables were examined via Chi-squared tests or Fisher’s exact tests, depending on their suitability. Multivariable logistic regression employing a stepwise covariate selection approach was conducted to discern statistically significant predictors. Following this analytical phase, the diagnostic accuracy of the constructed models was evaluated using receiver operating characteristic (ROC) curve analysis, with quantification of three key metrics: the area under the ROC curve (AUC) as a measure of discriminative capacity, along with diagnostic sensitivity and specificity parameters. The DeLong method was used to evaluate and compare the AUCs of between the qualitative and quantitative approaches.

Intraclass kappa coefficients were calculated from two repeated measurements by Q.F., while interclass kappa coefficients were derived from measurements by Q.F. and K.S. Following established interpretation criteria (18), kappa values were categorized as follows:

• <0: no agreement;
• 0–0.20: slight agreement;
• 0.21–0.40: fair agreement;
• 0.41–0.60: moderate agreement;
• 0.61–0.80: substantial agreement;
• 0.81–1.00: almost perfect agreement.

Statistical significance was defined as P<0.05, with adjustments for multiple comparisons performed using the Bonferroni method. All analyses were conducted in SPSS (v26.0, IBM) and R (v4.1.2).

Results

Patient characteristics

In this cohort of 98 patients, MRI revealed 98 papillary neoplasms exhibiting NME. Out of the 98 lesions, 52 were benign and 46 were malignant. The benign papillary lesions included intraductal papilloma (IDP) (n=47) and IDP with atypical ductal hyperplasia (ADH) (n=5). The malignant papillary lesions consisted of papillary ductal carcinoma in situ (DCIS) (n=2), encapsulated papillary carcinoma (EPC) (n=2), and solid papillary carcinoma (SPC) (n=42). Demographic characteristics, clinical manifestation, and mammography features of papillary neoplasms were presented in Table 1. It was found that there were significant differences in age, menopausal status, clinical manifestation, and mammography characteristics between patients with benign and malignant papillary lesions (P=0.005, <0.001, 0.01, 0.02, respectively).

Qualitative MRI assessment

Table 2 demonstrates the univariate analysis of qualitative MRI assessments for identifying malignant lesions. No statistically significant difference was observed in internal enhancement patterns between benign and malignant NME papillary lesions (P=0.10). However, there were significant differences in the amount of fibroglandular tissue (FGT), NME distribution, and time-intensity curve (TIC) pattern between the two groups (P=0.04, 0.008, <0.001, respectively). Variables demonstrating statistically significant associations in univariate analysis were included in the multivariate logistic regression model. The qualitative model yielded an AUC of 0.792 [95% confidence interval (CI): 0.701, 0.884], a sensitivity of 67.4%, and a specificity of 84.6%. The TIC pattern was found to be the independent predictor for diagnosing malignant NME papillary tumors. Although the initial phase of TIC is also important, only the delayed phase was examined.

Quantitative MRI histogram and texture analysis

LASSO identified 17 features, which were further narrowed down to 7 through stepwise logistic regression, as shown in Table 3. Compared with benign NME papillary neoplasms, malignant ones showed significant higher ADC_SD, ADC_diff-entropy, ADC_contrast, and SUB MR_{95th percentile} (P=0.009, 0.001, 0.01, 0.02, respectively), and significant lower ADC_{5th percentile}, DWI_b50-SD, and DWI_b800-mean (P=0.01, 0.001, 0.002, respectively). The quantitative model achieved an AUC of 0.908 (95% CI: 0.851, 0.966), a sensitivity of 82.6%, and a specificity of 88.5%. Figure 3 provided examples that illustrate the aforementioned results.

Figure 3 The histogram of benign and malignant papillary neoplasms. (A-H) A 64-year-old female with IDP; (I-P) a 57-year-old female with SPC. (A,C,I,K) DWI with b values of 50 and 800 s/mm²; (E,M) ADC map; (G,O) contrast-enhanced T1-weighted SUB MR images overlaid with color maps; (B,D,F,H,J,L,N,P) generated corresponding histogram of whole-lesion. ADC, apparent diffusion coefficient; DWI, diffusion-weighted imaging; IDP, intraductal papilloma; MR, magnetic resonance; SPC, solid papillary carcinoma; SUB, subtraction.

The quantitative model increased the sensitivity (67.4% vs. 82.6%) and specificity (84.6% vs. 88.5%) of the qualitative model. The AUC also improved (0.792 vs. 0.908, P<0.001), as shown in Figure 4.

Figure 4 ROC curves of qualitative MR assessment and quantitative whole-lesion histogram and texture analysis in distinguishing between benign and malignant papillary breast neoplasms on MRI with NME. MR, magnetic resonance; MRI, magnetic resonance imaging; NME, non-mass enhancement; ROC, receiver operating characteristic.

The AUC of the quantitative model for distinguishing malignant NME papillary neoplasms from benign NME papillary neoplasms in the internal validation set was 0.941, a sensitivity of 90.4%, and a specificity of 87.0%, as shown in Figure 5. The quantitative model results have been confirmed to be reliable through 10-fold cross-validation.

Figure 5 ROC curves of quantitative whole-lesion histogram and texture analysis model in internal validation set. AUC, area under the ROC curve; ROC, receiver operating characteristic.

Intra- and inter‑observer agreement

In all measurements of the whole-lesion histogram and texture analysis, we observed almost perfect intra-observer and inter-observer reliability, with intra- and inter-class coefficient values ranging from 0.881 to 0.989 and 0.809 to 0.955, respectively. Details are shown in Table 4.

Discussion

When it comes to detecting and diagnosing NME, it is widely acknowledged that this poses a more complex challenge compared to mass enhancement. A potential solution lies in adopting a radiomics approach. Our findings illustrated that the characteristics derived from quantitative histogram and texture analysis exhibit significantly higher AUC values when differentiating papillary neoplasms on MRI with NME, in comparison to the 2013 version of BI-RADS-MRI qualitative features. The features derived from DWI, ADC map, and SUB MR could serve as potential biomarkers for distinguishing malignant from benign lesions. As far as we are aware, this study represents the first investigation employ both qualitative assessments and quantitative histogram and texture analyses in evaluating the diagnostic accuracy of differentiating NME papillary neoplasms.

Our results show that the TIC pattern plays a crucial role in distinguishing between benign and malignant NME lesions, aligning with Li’s study (5). However, unlike Zhou’s research (19), where they identified the internal enhancement pattern as the sole statistically significant factor. This lack of disparity might be due to their smaller sample size. We intend to expand our research by increasing the sample size in future studies.

Research (20) has demonstrated that relying solely on the average ADC (potentially with just two b values) can significantly enhance the specificity of conventional contrast-enhanced breast MRI and, in turn, reduce the need for unnecessary biopsies. In our study, we computed ADC values using only two b values, revealing that malignant NME lesions exhibit lower 5^th percentile of ADC values. An earlier investigation (21) noted that minimal ADC values could serve as potential indicators of invasive elements within DCIS. Moreover, a different study’s (7) multivariable logistic regression analysis indicated that minimal ADC stood out as the sole independent predictor of breast malignancy. The minimum ADC value proves valuable in distinguishing between benign and malignant lesions more accurately than the mean ADC value because it represents the most diffusion-restricted region within a heterogeneous tumor. However, it’s worth noting that the minimum ADC value could be influenced by outliers stemming from noise, artifacts, and adjacent structures. Hence, the 5^th percentile value might offer a more accurate reflection of the actual scenario, although this requires further validation.

We chose to subtract the initial dynamic enhancement phase and the mask as these can more accurately represent the lesion’s early rapid enhancement. The greater the increase in signal intensity, the more extensive the portion of the tumor with rapid early enhancement, which is indicative of a higher level of biological aggression in the cancer (9). Our study found that the 95^th percentile of the SUB MR-related parameters of malignant lesions were significantly higher than those of benign ones, supporting the conclusion that greater early enhancement is associated with greater biological aggression in cancer.

In our study, the diff-entropy of the parameters related to ADC was significantly higher in malignant papillary NME lesions than in benign ones, which means that the gray levels randomness and entropy difference of the above sequences in benign papillary NME lesions are more notable than those in benign ones. Since malignant papillary NME lesions have worse prognosis and warrant different treatments (1), entropy-based features could be a useful textural biomarker for monitoring the progress of the disease. Waugh et al. (13) showed that entropy-based features are the most reliable indicators for distinguishing between different histological and molecular subtypes of breast cancer. Meanwhile, Fan et al. (22) revealed that luminal A tumors exhibited higher values of difference entropy feature compared to luminal B tumors. Other studies have demonstrated that entropy can effectively distinguish malignant from benign lesions (7,10). Chamming’s et al. (23) have proposed that alterations in T2-weighted imaging entropy following three cycles of neoadjuvant chemotherapy are more sensitive indicators of pathologic complete response compared to changes in tumor size after treatment. In terms of the correlation between entropy and prognosis, Kim et al. (9) discovered that increased entropy in T2-weighted MR images and reduced entropy in SUB MR images were linked to worse outcomes. Our findings, along with those of previous research, suggest that heterogeneity measured on breast MR images through entropy-based features can serve as a prognostic factor.

Recent studies have predominantly suggested that active surveillance may represent an appropriate management strategy for benign papillomas diagnosed via core needle biopsy, thereby potentially avoiding unnecessary overtreatment (24-26). Management of malignant papillary neoplasms in situ should follow treatment recommendations for DCIS (27-29). In cases with associated IDC, management prioritizes the treatment of the invasive component (27). Given the distinct clinical management approaches for benign and malignant papillary neoplasms, preoperative differentiation between these lesions is essential. CNB is considered not that accurate in NME because of the limited biopsy sampling, especially for the diffuse NME. Radiomics approach proves to be significant to differentiate the benign and malignant NME papillary neoplasms. So, radiomics approach can replace unnecessary biopsies for noninvasive preoperative diagnosis and assessment of the management of NME papillary neoplasms.

There are several strengths in our study. Firstly, our findings indicate that utilizing histogram and texture analysis of MR multiparametric maps may serve as a promising imaging surrogate for characterizing the benign and malignant breast papillary neoplasms on MRI with NME, to avoid overtreatment for benign lesions. Secondly, employing semi-automated segmentation and volumetric analysis helps minimize the inconsistencies associated with manual single-slice or multi-slice measurements, thus enhancing the reliability of results (30). Lastly, the use of histogram- and texture-based image features not only makes it easier for physicians with limited mathematical knowledge to understand and interpret the data, but also enables less experienced physicians to identify lesions more accurately.

Nevertheless, our research has certain limitations that warrant further investigation. Firstly, the study was conducted retrospectively and was based on a single MR system, which limits the generalizability of our findings. Secondly, we conducted the imaging at 1.5 T, but for upcoming studies, we could attain even greater DWI resolution, with thinner slices, by using higher field strengths, which would allow us to reduce partial volume effects and achieve a higher contrast-to-noise ratio, thereby enabling us to more accurately identify small or diffuse lesions. Thirdly, we only included DWI images with two b-values. However, research has shown that the DKI model has better diagnostic performance in distinguishing benign from malignant breast tumors compared to the traditional DWI model (8). Therefore, future research should focus on selecting DWI images with multiple b-values to verify and supplement our results. The texture features employed in this study were limited in complexity. Future investigations could benefit from incorporating a broader range of texture descriptors to enhance analysis.

Conclusions

In comparison to the qualitative BI-RADS assessment, utilizing MR multiparametric maps for whole lesion histogram and texture analysis can enhance the ability to differentiate between benign and malignant NME papillary breast lesions, in order to avoid overtreatment. Nonetheless, more extensive controlled trials are required to fully confirm the clinical efficacy of this method.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-128/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-128/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-128/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-128/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of Ruijin Hospital (Shanghai Jiao Tong University School of Medicine) [No. 2023(333)] and individual consent for this retrospective analysis was waived due to the retrospective nature.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Li X, Febres-Aldana C, Zhang H, et al. Updates on Lobular Neoplasms and Papillary, Adenomyoepithelial, and Fibroepithelial Lesions of the Breast. Arch Pathol Lab Med 2022;146:930-9. [Crossref] [PubMed]
2. Kulka J, Madaras L, Floris G, et al. Papillary lesions of the breast. Virchows Arch 2022;480:65-84. [Crossref] [PubMed]
3. Qiu L, Mais DD, Nicolas M, et al. Diagnosis of Papillary Breast Lesions on Core Needle Biopsy: Upgrade Rates and Interobserver Variability. Int J Surg Pathol 2019;27:736-43. [Crossref] [PubMed]
4. Jatana S, Zahra S, Miron AD, et al. A retrospective study evaluating outcomes of surgical excision versus radiologic follow-up of benign breast papillomas without atypia. Am J Surg 2022;224:932-7. [Crossref] [PubMed]
5. Li Y, Yang ZL, Lv WZ, et al. Non-Mass Enhancements on DCE-MRI: Development and Validation of a Radiomics-Based Signature for Breast Cancer Diagnoses. Front Oncol 2021;11:738330. [Crossref] [PubMed]
6. Vidić I, Egnell L, Jerome NP, et al. Support vector machine for breast cancer classification using diffusion-weighted MRI histogram features: Preliminary study. J Magn Reson Imaging 2018;47:1205-16. [Crossref] [PubMed]
7. Liu W, Wei C, Bai J, et al. Histogram analysis of diffusion kurtosis imaging in the differentiation of malignant from benign breast lesions. Eur J Radiol 2019;117:156-63. [Crossref] [PubMed]
8. Li T, Hong Y, Kong D, et al. Histogram analysis of diffusion kurtosis imaging based on whole-volume images of breast lesions. J Magn Reson Imaging 2020;51:627-34. [Crossref] [PubMed]
9. Kim JH, Ko ES, Lim Y, et al. Breast Cancer Heterogeneity: MR Imaging Texture Analysis and Survival Outcomes. Radiology 2017;282:665-75. [Crossref] [PubMed]
10. Li X, Jiang N, Zhang C, et al. Value of conventional magnetic resonance imaging texture analysis in the differential diagnosis of benign and borderline/malignant phyllodes tumors of the breast. Cancer Imaging 2021;21:29.
11. Eun NL, Kang D, Son EJ, et al. Texture analysis using machine learning-based 3-T magnetic resonance imaging for predicting recurrence in breast cancer patients treated with neoadjuvant chemotherapy. Eur Radiol 2021;31:6916-28. [Crossref] [PubMed]
12. Eun NL, Kang D, Son EJ, et al. Texture Analysis with 3.0-T MRI for Association of Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology 2020;294:31-41. [Crossref] [PubMed]
13. Waugh SA, Purdie CA, Jordan LB, et al. Magnetic resonance imaging texture analysis classification of primary breast cancer. Eur Radiol 2016;26:322-30. [Crossref] [PubMed]
14. Xie T, Zhao Q, Fu C, et al. Differentiation of triple-negative breast cancer from other subtypes through whole-tumor histogram analysis on multiparametric MR imaging. Eur Radiol 2019;29:2535-44. [Crossref] [PubMed]
15. Sun K, Zhu H, Chai W, et al. Whole-lesion histogram and texture analyses of breast lesions on inline quantitative DCE mapping with CAIPIRINHA-Dixon-TWIST-VIBE. Eur Radiol 2020;30:57-65. [Crossref] [PubMed]
16. Qin Y, Tang C, Hu Q, et al. Assessment of Prognostic Factors and Molecular Subtypes of Breast Cancer With a Continuous-Time Random-Walk MR Diffusion Model: Using Whole Tumor Histogram Analysis. J Magn Reson Imaging 2023;58:93-105. [Crossref] [PubMed]
17. American College of Radiology. ACR BI-RADS® Magnetic Resonance Imaging 2013. In: Morris EA, Comstock C, Lee C, et al. ACR BI-RADS Atlas: Breast Imaging Reporting and Data System. 5th ed. Reston: American College of Radiology; 2013.
18. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-74.
19. Zhou J, Li M, Liu D, et al. Differential Diagnosis of Benign and Malignant Breast Papillary Neoplasms on MRI With Non-mass Enhancement. Acad Radiol 2023;30:S127-32. [Crossref] [PubMed]
20. McDonald ES, Romanoff J, Rahbar H, et al. Mean Apparent Diffusion Coefficient Is a Sufficient Conventional Diffusion-weighted MRI Metric to Improve Breast MRI Diagnostic Performance: Results from the ECOG-ACRIN Cancer Research Group A6702 Diffusion Imaging Trial. Radiology 2021;298:60-70. [Crossref] [PubMed]
21. Kim JY, Kim JJ, Lee JW, et al. Risk stratification of ductal carcinoma in situ using whole-lesion histogram analysis of the apparent diffusion coefficient. Eur Radiol 2019;29:485-93. [Crossref] [PubMed]
22. Fan M, Cheng H, Zhang P, et al. DCE-MRI texture analysis with tumor subregion partitioning for predicting Ki-67 status of estrogen receptor-positive breast cancers. J Magn Reson Imaging 2018;48:237-47. [Crossref] [PubMed]
23. Chamming's F, Ueno Y, Ferré R, et al. Features from Computerized Texture Analysis of Breast Cancers at Pretreatment MR Imaging Are Associated with Response to Neoadjuvant Chemotherapy. Radiology 2018;286:412-20. [Crossref] [PubMed]
24. Elfgen C, Leo C, Kubik-Huch RA, et al. Third International Consensus Conference on lesions of uncertain malignant potential in the breast (B3 lesions). Virchows Arch 2023;483:5-20. [Crossref] [PubMed]
25. Rubio IT, Wyld L, Marotti L, et al. European guidelines for the diagnosis, treatment and follow-up of breast lesions with uncertain malignant potential (B3 lesions) developed jointly by EUSOMA, EUSOBI, ESP (BWG) and ESSO. Eur J Surg Oncol 2024;50:107292. Erratum in: Eur J Surg Oncol 2024;50:107943. [Crossref] [PubMed]
26. Rageth CJ, O'Flynn EAM, Pinker K, et al. Second International Consensus Conference on lesions of uncertain malignant potential in the breast (B3 lesions). Breast Cancer Res Treat 2019;174:279-96. [Crossref] [PubMed]
27. Chen P, Zhou D, Wang C, et al. Treatment and Outcome of 341 Papillary Breast Lesions. World J Surg 2019;43:2477-82. [Crossref] [PubMed]
28. Tan HJ, Tan PH, Leong LCH, et al. Encapsulated papillary carcinoma of the breast: An institutional case series and literature review. Cancer Med 2023;12:11408-16. [Crossref] [PubMed]
29. Changrong W, Zhibo Z, Jufeng G, et al. Encapsulated Papillary Carcinoma of the Breast: A Review of Clinicopathologic Characteristics, Molecular Mechanisms, and Patient Management. Cancer Control 2024;31:10732748241299071. [Crossref] [PubMed]
30. Barwick T, Orton M, Koh DM, et al. Repeatability and reproducibility of apparent diffusion coefficient and fat fraction measurement of focal myeloma lesions on whole body magnetic resonance imaging. Br J Radiol 2021;94:20200682. [Crossref] [PubMed]