Breast Cancer Subtypes and Quantitative Magnetic Resonance Imaging: A Systemic Review

Magnetic resonance imaging (MRI) is the most sensitive imaging modality for breast cancer detection. This systematic review investigated the role of quantitative MRI features in classifying molecular subtypes of breast cancer. We performed a literature search of articles published on the application of quantitative MRI features in invasive breast cancer molecular subtype classification in PubMed from 1 January 2002 to 30 September 2021. Of the 1275 studies identified, 106 studies with a total of 12,989 patients fulfilled the inclusion criteria. Bias was assessed based using the Quality Assessment of Diagnostic Studies. All studies were case-controlled and research-based. Most studies assessed quantitative MRI features using dynamic contrast-enhanced (DCE) kinetic features and apparent diffusion coefficient (ADC) values. We present a summary of the quantitative MRI features and their correlations with breast cancer subtypes. In DCE studies, conflicting results have been reported; therefore, we performed a meta-analysis. Significant differences in the time intensity curve patterns were observed between receptor statuses. In 10 studies, including a total of 1276 lesions, the pooled difference in proportions of type Ⅲ curves (wash-out) between oestrogen receptor-positive and -negative cancers was not significant (95% confidence interval (CI): [−0.10, 0.03]). In nine studies, including a total of 1070 lesions, the pooled difference in proportions of type 3 curves between human epidermal growth factor receptor 2-positive and -negative cancers was significant (95% CI: [0.01, 0.14]). In six studies including a total of 622 lesions, the pooled difference in proportions of type 3 curves between the high and low Ki-67 groups was significant (95% CI: [0.17, 0.44]). However, the type 3 curve itself is a nonspecific finding in breast cancer. Many studies have examined the relationship between mean ADC and breast cancer subtypes; however, the ADC values overlapped significantly between subtypes. The heterogeneity of ADC using kurtosis or difference, diffusion tensor imaging parameters, and relaxation time was reported recently with promising results; however, current evidence is limited, and further studies are required to explore these potential applications.


Introduction
Breast cancer is the most frequently diagnosed malignancy and the leading cause of cancer-related deaths among women [1]. Breast cancer is a heterogeneous disease with a high degree of diversity in the risks of therapeutic resistance and disease progression [2,3]. Therefore, individualized management is widely accepted [2,3]. However, previous classifications based on tumour size, grade, and histology cannot completely reflect tumour characteristics. Gene expression profiling has revealed four major breast cancer subtypes: luminal-A, luminal-B, human epidermal growth factor receptor 2 (HER2)-enriched, and basal-like [3]. Each subtype has varied prognoses, progression risks, responses to treatment, and survival outcomes. Commercial multigene assays are expensive and time-consuming; from 1 January 2002 to 30 September 2021. One reviewer performed the data acquisition using the following search terms: "magnetic resonance imaging", "breast neoplasms" and "subtype or phenotype or Ki-67 or receptors, oestrogen or receptors, progesterone or receptor, and ErbB-2". The secondary references were manually checked and included in the study.

Inclusion and Exclusion Criteria
The primary endpoint of the systematic review was the association between molecular subtypes of breast cancer and quantitative MRI values. Studies (or subsets of studies) were included if they satisfied all the following criteria: (1) inclusion of patients with invasive breast cancer confirmed by histopathology, (2) pre-treatment MRI, (3) quantitative analysis of MRI, (4) MRI correlation with breast cancer subtypes or factors that determine subtype, (5) human women, and (6) English language. We considered studies reporting visual evaluations, such as high signal intensity and heterogeneity without quantification, to be qualitative studies and did not include them. The exclusion criteria were: (1) reviews without meta-analyses, case reports, or editorials, and (2) radiomics, machine learning, or artificial intelligence studies. We regarded simple methods, such as histogram analysis, diffusion tensor imaging (DTI), and pharmacokinetic analysis, as non-radiomics. The results of this study are summarised in Figure 1. Ethical approval was not required for this study.

Article Selection and Data Extraction
One radiologist screened the selected titles and abstracts to ensure conformity with the inclusion criteria and documented the rationale for exclusion. Supplementation for article selection was performed by screening the reference lists. After screening, the full texts were reviewed. The following data were extracted from the literature: authors, year of publication, number of patients, number of cases in each subtype, sequences and analytic methods included in image analysis, and results.

Article Selection and Data Extraction
One radiologist screened the selected titles and abstracts to ensure conformity with the inclusion criteria and documented the rationale for exclusion. Supplementation for article selection was performed by screening the reference lists. After screening, the full texts were reviewed. The following data were extracted from the literature: authors, year of publication, number of patients, number of cases in each subtype, sequences and analytic methods included in image analysis, and results.

Quality Assessment
The methodological quality of the acquired studies was checked based on the Quality Assessment of Diagnostic Studies (QUADAS 2) [17].

Data Synthesis
When there were more than five studies with similar methodologies, conflicting results, and no prior meta-analysis, a meta-analysis was performed using RevMan v5.4 (Cochrane Collaboration, London, UK). The mean difference in the prevalence of imaging findings was analysed using a random effects model. In the analysis of the apparent diffusion coefficient (ADC), we suspected that region-of-interest (ROI) placement might be a cause of heterogeneous results; therefore, we identified the ROI placement methods and classified them.

Literature Search
A total of 1267 articles were identified in the electronic databases. Following the removal of 485 duplicates, the titles and abstracts of 782 articles were screened. Of the 782 articles, 596 did not fulfil the inclusion criteria. Nine laboratory studies, 23 qualitative studies, 16 reviews or case reports, and 40 studies with radiomics were excluded. Eight studies were identified from the citation search and were also included. Figure 1 shows an overview of the literature search and study selection process.

Study Characteristics
The included studies encompassed 104 original studies and two meta-analyses, with publication years ranging from 2003 to 2021. A total of 12,989 patients, excluding doublecounted patients in prior meta-analyses (n = 3466), met the selection criteria. Quantitative features were mostly derived from kinetic parameters measured using DCE (n = 45) or diffusion-weighted imaging (DWI) (n = 68). Five studies assessed relaxation time and four assessed magnetic resonance spectroscopy (MRS). Of these, 10 studies analysed both DWI and DCE, one DCE and relaxation time, one DCE and MRS, one DCE, DWI, and relaxation time, and one DCE, DWI, and MRS.

Methodological Quality of the Included Studies
Patient selection was generally well-defined within the respective methodology. However, in 18 studies, more than 10% of cases were excluded for ambiguous reasons such as poor image quality. In one study, the sum of cases did not match the total number. This may have contributed to potential bias. All studies reported the methodology of the index test and were, thus, not considered a significant source of potential bias. Although immunohistochemical staining criteria differed among studies, the reference standards in all studies were histopathology with immunohistochemical staining and were not considered a significant source of potential bias. The subtype classification method was adapted from the 2011 St. Gallen Consensus meeting [18]. All patients underwent the reference test with the appropriate timing when they were included in the analysis.

Dynamic Contrast-Enhanced-Magnetic Resonance Imaging
Investigations regarding DCE-MRI are summarised in Table 1. DCE-MRI offers information not only on lesion cross-sectional morphology, but also on functional lesion features, such as tissue perfusion and enhancement kinetics [14]. In DCE-MRI, highly vascularised tumours tend to show early and strong contrast enhancement and wash-out of contrast in the delayed phase. Many methods have been proposed, with the proportion of time-intensity curve patterns (n = 14) being the most widely used.   Twelve studies evaluated time-intensity curve patterns and ER status; however, two studies did not show the exact values and were excluded from this meta-analysis [50,53]. In the meta-analysis of these 10 studies, including a total of 1276 lesions [19][20][21][23][24][25]30,43,46,47], the pooled difference in proportions of type III curves (wash-out) between ER-positive cancer and ER-negative cancers for all included tumours was −0.04, (95% confidence interval [CI] = [−0.10, 0.03]), heterogeneity τ 2 = 0.00, I 2 = 39%, test for overall effect Z = 1.13 (p = 0.26) (Figure 2a).
Twelve studies evaluated time-intensity curve patterns and ER status; however, two studies did not show the exact values and were excluded from this meta-analysis [50,53]. In the meta-analysis of these 10 studies, including a total of 1276 lesions [19][20][21][23][24][25]30,43,46,47], the pooled difference in proportions of type Ⅲ curves (wash-out) between ER-positive cancer and ER-negative cancers for all included tumours was −0.04, (95% confidence interval [CI] = [−0.10, 0.03]), heterogeneity τ 2 = 0.00, I 2 = 39%, test for overall effect Z = 1.13 (p = 0.26) (Figure 2a). A pharmacokinetic analysis was performed in 11 studies. Ktrans is a transfer constant that measures the rate of transport of contrast medium from the plasma to the extravascular extracellular space (EES), and provides a measure of vascular permeability and blood flow. Ve is the tumour volume occupied by the EES and Kep describes the outflow rate of the contrast medium from the EES back to the plasma. Higher Kep and lower Ve values in DCE-MRI were observed in the TN subtype [27,28]. Two studies analysed the relationship between TN cancers and pharmacokinetic parameters. Both reported significantly higher Kep and lower Ve values in TN cancers than in other subtypes [27,28]. However, the relationships between other pharmacokinetic parameters and prognostic factors were conflicting. Six studies evaluated whether HER2-positive cancers had a higher Ktrans than that in HER2-negative cancers. Two of them demonstrated significant differences [38,61], and the other four demonstrated no significant differences [28,37,39,48]. Similarly, eight studies evaluated the relationship between Ki-67 status and Ktrans. Three studies demonstrated significant differences, and five studies demonstrated no significant differences [28,34,[37][38][39]48,56,61]. These studies used the same model proposed by Tofts [65]; however, there were highly variable values between the studies, which hindered the meta-analysis. For example, the mean Ktrans of invasive breast cancers with a low Ki-67 (Ki-67 < 14 %) in one study was 2.56/min [61], whereas that in another study was 0.18/min [38].
Two studies reported that the HER2 subtype exhibited higher rapid early contrast uptake [36,42]. Many other indices have been proposed; however, these have been evaluated in only a few studies or conflicting results were reported. For example, two studies reported that a short peak time was associated with positive HER2 status [19,24]; however, three did not find any significant differences [20,30,32]. Three studies evaluated background parenchymal enhancement (BPE) and breast cancer subtypes [54,55,63]. One study reported that moderate and marked BPE prevailed over minimal and mild BPE in patients with TN cancers [54], whereas another reported that BPE was significantly lower in patients with TN cancer compared with patients with non-TN cancers [63].
The DWI results are summarised in Table 2. Sixty-three of the 68 DWI studies analysed the ADC. There have been two meta-analyses regarding the subtypes and Ki-67 [66,67]. Meyers et al. reported that the ADC values of breast cancer subtypes overlapped significantly, with no clear proposed threshold to distinguish between them [66]. In this metaanalysis, the I 2 ranged from 95% to 98%, suggesting considerable heterogeneity. Surov      These meta-analyses may be affected by heterogeneous methodologies, one of which might be the ROI placement. We classified ROI placement as follows: evaluation of the whole lesion (whole), solid portion of the lesion excluding cystic/necrotic/haemorrhagic portion (solid), and ROI placement methodology not found (unknown). One study evaluated the effect of ROI placement using both solid and whole methods [108]. Seventeen studies with a known ROI placement methodology evaluated whether TN breast cancers had higher ADC values than other cancer subtypes. In the solid portion measurement group, one of the 12 studies demonstrated significant differences [ four of the five studies reported significant differences in the whole lesion measurement group [29,51,78,80,96]. Twelve studies with known ROI placement methodology evaluated whether luminal-B-type breast cancers had lower ADC values than luminal-A-type cancers. In the solid portion measurement group, six out of nine studies demonstrated significant differences [41,50,77,79,85,97,105,110,117], while one out of three studies demonstrated significant differences in the whole lesion measurement group [78,80,98].
Although ADC values differed among breast cancer subtypes, the ADC values of different tumour subtypes overlapped significantly [66,110]. Instead of using the mean ADC, more sophisticated methods, such as differences in ADC and diffusion kurtosis, have been evaluated with promising results. Two studies evaluated the relationship between ADC differences (maximum ADC to minimum ADC) and subtypes. They reported that the ADC difference was significantly associated with Ki-67 expression [98,111].
In probability theory and statistics, the alteration of a normative distribution pattern is known as kurtosis. Diffusion kurtosis imaging attempts to account for this variation to provide a more accurate model of diffusion as a reflective marker for tissue heterogeneity [119]. Similarly, skewness, which reflects the asymmetry of ADC value distribution, has been introduced in cancer imaging [105,118]. Three studies reported a positive association between diffusion kurtosis and the Ki-67 index [81,107,109]. Similarly, one study reported significantly higher ADC kurtosis in the TN group than in the ER-positive group [96].

Intravoxel Incoherent Motion
Intravoxel incoherent motion (IVIM) MRI is a non-invasive imaging method that allows the evaluation of both tissue diffusivity and tissue microcapillary perfusion. When DWI is performed with multiple b-values (usually 0-1000 smm −2 ), the signal intensity at low b-values (e.g., 0-100 smm −2 ) reflects both water diffusion in tissues and microcirculation within the capillaries. In contrast, at higher b-values, the signal intensity is more reflective of tissue diffusivity. Thus, the classical IVIM model uses a biexponential analysis that provides the tissue diffusion coefficient (Dt), perfusion-related diffusion (Df), and perfusion fraction (f).
Ten studies evaluated the association between IVIM and breast cancer subtypes. Seven studies evaluated whether high-Ki-67 tumours had lower Dt than low-Ki-67 tumours. Three of them demonstrated significant differences [41,89,104], whereas the other four demonstrated no significant differences [8,64,93,115]. Five studies evaluated whether HER2positive cancers had higher Df than HER2-negative cancers. Two demonstrated significant differences [89,115], and the other three demonstrated no significant differences [8,52,93]. Owing to the large heterogeneity of the results, we did not perform a meta-analysis.

Diffusion Tensor Imaging
DTI is a conceptual framework that provides quantitative information on the directional diffusivity of water molecules [120]. The measurement of DTI indices, such as ADC, fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), geodesic anisotropy (GA), relative anisotropy (RA), and volume ratio (VR), provides quantification. The mammary ductal network may result in diffusion anisotropy in healthy fibroglandular tissue [121]; however, cancer cells may destroy these structures, leading to reduced anisotropy. Two studies reported that FA was significantly higher in the low-Ki-67 group and ER-positive cancers [92,94].

Relaxation Time
The relaxation time findings are summarised in Table 3. One study reported significantly longer T2* relaxation times in higher histologic grades, which correlated with high signal intensity on T2-weighted imaging [122]. Using synthetic MRI, three studies assessed T1 and T2 relaxation times [57,61,123]. Two reported significantly higher T2 in the HR-negative group compared to the HR-positive group [61,123].

Magnetic Resonance Spectroscopy
The MRS findings are summarised in Table 4. MRS provides biochemical information regarding the investigated tissues. Increased choline (Cho) is a marker of elevated cellular proliferation rates in breast cancer [125]. Four studies evaluated the relationship between MRS and subtypes [21,50,126,127]. Conflicting results were reported with TN breast cancers and MRS [50,126].

DCE-MRI
DCE-MRI is a standard diagnostic technique with high sensitivity and variable specificity for characterising breast lesions. Angiogenesis is one of the main factors affecting gadolinium uptake and contributes to internal enhancement patterns and kinetic curves. In DCE-MRI, highly vascularised tumours tend to show early and strong contrast enhancement and wash-out of contrast in the delayed phase. This study demonstrated that a significantly higher proportion of type 3 curves was observed in the high-Ki-67 group compared with the low-Ki-67 group. This finding was consistent with the correlation between vascular endothelial growth factor (VEGF) and histological grade reported in a pathological study [128]. In the meta-analysis, a significantly higher proportion of type 3 curves was observed in HER2-positive cancers than in HER2-negative cancers. This finding was consistent with the correlation between the overexpression of VEGF and HER2-positive tumours in pathological studies [6][7][8]22,36].
Although a negative correlation between ER status and cytosolic levels of VEGF has been reported in pathological studies [128] and the proportion of type 3 curves tended to be lower in ER-positive cancers, no significant difference was observed in this meta-analysis.
However, the wash-out curve itself is a common finding in breast cancer, and the prediction of subtypes based on this finding is difficult. Many indices have been proposed; however, other indices are immature and conflicting results have been reported.

DWI
DWI detects the Brownian motion of water protons, thereby reflecting the biological characteristics of the tissue. ADC is used to quantify Brownian motion. By imaging alterations in the microscopic motion of water molecules, DWI can yield novel quantitative and qualitative information reflecting cellular changes that can provide unique insights into tumour cellularity, with a potential role in the characterisation of breast masses [129]. The decreased ADC values in malignant tumours may be due to their increased cellularity, larger nuclei with more abundant macromolecular proteins, and reduced extracellular space. These tissue factors hinder proton diffusion and, consequently, lower ADC values [66,130].
Higher Ki-67 expression usually implies rapid proliferation, and consequently, increased cellularity, which restricts the diffusion of water molecules in the extracellular and extravascular spaces and is presumed to cause reduced ADC values [131]. A weak inverse correlation between tumour cellularity and ADC values has been described, and further associations between proliferation rate and tumour aggressiveness have been proposed [67,77,129].
However, several studies have suggested that highly aggressive invasive breast cancers rapidly outgrow their vascular supply in certain areas, leading to prolonged hypoxia within the tumour and subsequent necrosis [106,[132][133][134]. Areas of intratumoral necrotic tissue and loss of cell membrane integrity are associated with increased intratumoral water diffusion. This may explain the higher ADC value in the TN subtype when the entire lesion ADC is measured [78,135,136].
Neoangiogenesis is the basis of cancer cell proliferation. Pathological studies have demonstrated an association between cytosolic levels of VEGF, an angiogenesis stimulator, and histologic grade, as well as a negative correlation with ER status [6][7][8]22,36,128]. Owing to the perfusion effect, high vascularity can result in increased ADC values. Furthermore, tumour vessels tend to have larger diameters than normal microvessels as well as discontinuities in the vascular walls, leading to increased total extracellular fluid volumes. The higher tumour blood flow and increased extracellular fluid appear to compensate for the low ADC of high cellularity [37,73,78,80,101,112,137,138].
These paradoxical phenomena may cause confusion in subtype predictions based on the mean ADC. Meyer et al. reported in a meta-analysis that ADC values cannot discriminate immunohistochemical molecular subtypes [66]. To overcome the increased ADC by necrosis, methods of assessing heterogeneity using ADC kurtosis and ADC differences, which may reflect high cellular areas and necrosis, have been proposed, with promising results [63,96,98,107,111,118].
feature of breast cancer, which may reflect this fibrosis ( Figure 3) [14,1 TN breast cancers have high signal intensity on T2-weighted image [29,[144][145][146]. In addition, a higher tumour grade often correlates wi genesis [128]. Angiogenesis increases total extracellular fluid volume high-grade tumours may demonstrate high signal intensity on T2-we signal intensity on T2-weighted images is also correlated with tum 149]. These studies involved subjective qualitative analyses, which makes it difficult to apply their results in clinical practice to assess the HR status or subgroup categorisation of ER-positive breast cancers. Seo et al. reported significantly longer T2 * relaxation times in higher histological grades [122]. Recent advances in quantitative MRI have enabled the acquisition of both MR images and quantitative MR data in a single scan [57,150]. Synthetic MRI enables us to appreciate subtle quantitative MRI value differences that are invisible to radiologists' eyes alone. Synthetic MRI can also measure T1 values and proton density, which cannot be assessed using T2-weighted images [150][151][152]. Contrary to the ADC, T2 values of highly proliferative tumours were higher than those of low proliferative tumours [61,123]. Because these value assessments do not experience a paradoxical phenomenon, they may be more useful than ADC. In a radiomics study, Liang et al. reported that a T2-weighted image-based radiomics classifier was a significant predictor of Ki-67 status in patients with breast cancer, whereas contrast-enhanced image-based classifiers failed to discriminate in the validation dataset [153].

Luminal-Type Breast Cancer
This section explains the MRI characteristics of the breast cancer subtypes. In general, HR-positive tumours demonstrated stromal reaction, fibrosis, and perilesional spiculations [139]. An irregular mass margin and a non-round shape were significantly associated with luminal-A-type cancers [32,140,154]. Intratumoral iso/low T2-signal intensity may reflect fibrosis and is also associated with the luminal subtypes [14,[140][141][142]. Multifocal or multicentric carcinoma is less common in the luminal-A type than in the luminal-B or HER2 types [140,155]. Compared to the other subtypes, luminal-A-type breast cancers tended to show less strong enhancement [35]. Kato et al. reported that rim enhancement occurred significantly less frequently in luminal-A-type breast cancers [85] (Figure 3).
Tumour roundness is positively correlated with Ki-67 index [154]. Luminal-B subtypes are more often associated with multicentric/multifocal disease than are luminal-A cancers [32,156,157] and are also enriched for fibroblast growth factor receptor gene amplification, which has been implicated in angiogenesis [33,158]. This may lead to a higher ratio of lesion enhancement on DCE-MRI and heterogeneous internal enhancement [32,33,41,159] ( Figure 4). focal or multicentric carcinoma is less common in the luminal-A type tha B or HER2 types [140,155]. Compared to the other subtypes, luminal-A cers tended to show less strong enhancement [35]. Kato et al. reported t ment occurred significantly less frequently in luminal-A-type breast ca 3).
Tumour roundness is positively correlated with Ki-67 index [154] types are more often associated with multicentric/multifocal disease th cancers [32,156,157] and are also enriched for fibroblast growth factor r plification, which has been implicated in angiogenesis [33,158]. This ma ratio of lesion enhancement on DCE-MRI and heterogeneous intern [32,33,41,159] (Figure 4).

HER2-Enriched Subtype
HER2, a transmembrane receptor tyrosine kinase in the epidermal growth factor receptor family, is amplified or overexpressed in approximately 20% of breast cancers and is associated with poor prognosis, although it responds well to HER2-targeted therapies [4,6]. The cellular-level effects of HER2 overexpression include increased cell proliferation, cell survival, mobility, and invasiveness, as well as neo-angiogenesis by increasing VEGF production [6][7][8]. On gross pathology, a smooth mass margin was associated with the HER2-enriched subtype [32] (Figure 5). The presence of microcalcifications, especially branching or fine linear morphology, was associated with mammography [6]. HER2enriched subtypes are more often associated with multicentric/multifocal disease than luminal-A cancers [32,156,157,160]. Increased angiogenesis in the HER2-enriched subtype leads to rapid early contrast uptake and a higher proportion of wash-out curves on DCE-MRI ( Figure 2) [36].
the HER2-enriched subtype [32] (Figure 5). The presence of microcalcifications, es branching or fine linear morphology, was associated with mammography [6]. H riched subtypes are more often associated with multicentric/multifocal disease minal-A cancers [32,156,157,160]. Increased angiogenesis in the HER2-enriched leads to rapid early contrast uptake and a higher proportion of wash-out curves MRI (Figure 2) [36].

TN Breast Cancer
TN breast cancer is highly associated with the presence of a central scar, tum crosis, the presence of spindle cells or squamous metaplasia, high total mitotic co high nuclear-cytoplasmic ratio [9,10,145]. These cancers are also more likely round, oval, or lobulated masses and are more likely to be unifocal comp ER+/PR+/HER2 tumours [29,145,146,161,162]. MRI often shows areas of intratumo T2 signal intensity, lobulated shape, rim enhancement, and smooth margins (F

TN Breast Cancer
TN breast cancer is highly associated with the presence of a central scar, tumour necrosis, the presence of spindle cells or squamous metaplasia, high total mitotic count, and high nuclear-cytoplasmic ratio [9,10,145]. These cancers are also more likely to show round, oval, or lobulated masses and are more likely to be unifocal compared to ER+/PR+/HER2 tumours [29,145,146,161,162]. MRI often shows areas of intratumoral high T2 signal intensity, lobulated shape, rim enhancement, and smooth margins ( Figure 6) [29,[144][145][146]162]. The rim enhancement can be explained by high angiogenesis in the periphery of the tumour. Very high intratumoral signal intensity on T2-weighted MR images and an elongated T2 relaxation time may be associated with intratumoral necrosis [29,123,145]. When the necrotic areas are included, the ADCs of TN cancers are higher than of luminal-type breast cancers [29,51,78,96].  [29,[144][145][146]162]. The rim enhancement can be explained by high angiogenesis riphery of the tumour. Very high intratumoral signal intensity on T2-weighted M and an elongated T2 relaxation time may be associated with intratumora [29,123,145]. When the necrotic areas are included, the ADCs of TN cancers a than of luminal-type breast cancers [29,51,78,96].

Limitations
A major limitation of this review was the exclusion of complex radiomic Forty studies reporting radiomics in breast MRI and breast cancer subtypes were from this analysis ( Figure 1). Several radiomics methods have been proposed, w ising results. Further studies, including systemic reviews in this field, are warra meta-analysis only included studies published in English, resulting in selectio the data might not be representative of the non-native English-speaking regi world.

Conclusions
Conventional quantitative MRI features, such as the time-intensity curve ADC, might play a limited role in the prediction of breast cancer subtypes. W placement is essential for quantitative analysis, it currently depends on the ra Aggressive breast cancers, especially the TN subtype, contain necrosis, which c erogeneity within the tumour. Sophisticated evaluation of tumour heterogenei research of recently introduced techniques, and standardised interpretation of M may improve non-invasive breast cancer subtype classification and personal ment for patients with breast cancer.

Limitations
A major limitation of this review was the exclusion of complex radiomics studies. Forty studies reporting radiomics in breast MRI and breast cancer subtypes were excluded from this analysis (Figure 1). Several radiomics methods have been proposed, with promising results. Further studies, including systemic reviews in this field, are warranted. This meta-analysis only included studies published in English, resulting in selection bias, as the data might not be representative of the non-native English-speaking regions of the world.

Conclusions
Conventional quantitative MRI features, such as the time-intensity curve and mean ADC, might play a limited role in the prediction of breast cancer subtypes. While ROI placement is essential for quantitative analysis, it currently depends on the radiologists. Aggressive breast cancers, especially the TN subtype, contain necrosis, which causes heterogeneity within the tumour. Sophisticated evaluation of tumour heterogeneity, further research of recently introduced techniques, and standardised interpretation of MR images may improve non-invasive breast cancer subtype classification and personalised treatment for patients with breast cancer.