Table 1 Details of the studies included in this review

Gallivanone et al. [22]

Italy

To identify genetic tumor mechanisms and tumor phenotypes by radiogenomics in breast cancer

To combine phenotype data that are collected from imaging of tumors and genomics

Cohort

The Cancer Genome Atlas (TCGA) including the profiles of 233 BC Luminal A, 103 BC Luminal B, 43 BC HER2 + , and 74 BC basal patients

Comparison was made on the BC subtypes and 113 normal cases

A computational approach that relates the phenotype of extracted magnetic resonance imaging breast cancer (MRI BC) lesions to mRNAs and microRNAs (miRNAs)

Associated regulatory networks with the development of a radiomiRNomic map

Evaluated the relationship between MRI and miRNA in patients with breast cancer (BC)

Determined 16 radioactivity characteristics of the tumor phenotype

AUC:

BRCA1:0.6

BRCA2:0.5

Radiogenomics has more power in classification, differential diagnosis, and prognosis of BC

18

Grimm et al. [15]

UK

To recognize associations between extracted MRI features and the breast cancer molecular subtypes (BCMSTs)

Retrospective

275

preoperative breast MRIs

Evaluated MRI imaging from six trained breast imaging

Extracted 56 imaging morphology, dynamic, and tissue features

Classified tumors with molecular subgroups: human epidermal growth factor receptor 2 (HER2), basal, luminal A, and luminal B

–

This study showed the relationship between MSTs of luminal B and luminal A with imaging features

19

Zhang et al. [21]

China

To evaluate the effect of radiogenomics (Ki-67 + MRI) on the proliferation index in BC patients

Cohort

Invasive ductal breast cancer was identified in 128 clinicopathologically patients

32 negative Ki67 expressions versus 96 positive Ki67 expressions included

Before surgery, patients underwent diffusion-weighted MRI (DW-MRI)

Radiomics features were taken out from apparent diffusion coefficient (ADC) maps

• To forecast the Ki-67 index, the logistic regression (LR) model was recognized

AUC:0.75 ± 0.08, accuracy of 0.71

Ki-67 index was estimated as a noninvasive technique by three-dimensional images feature of ADC maps before surgery in BC

20

Park et al. [23]

Korea

To evaluate the effect of radiogenomics (ultrasound phenotypes in BC treatment and RNA sequencing revealed genomic changes)

Prospective

B-mode features like size, shape, echo, pattern, orientation, calcifications, and margin were examined on 31 women with breast cancer

By using contrast-enhanced ultrasound (CEUS) and microvascular ultrasound (US), vascular features were evaluated

By using next-generation sequencing, RNA sequencing was done with total RNA

Comparison between US feature and ingenuity pathway analysis and gene expression was made

US feature was compared to improved functions

–

B-mode and vascular US features in comparison with RNA sequencing revealed genomic changes related to angiogenesis, hormone receptor status, or diagnosis in BC

21

Yeh et al. [24]

USA

To evaluate radiomics' breast imaging using MRI related to gene pathways following RNA sequencing from invasive BCs prior to treatment

Retrospective

By implementing total RNA, around 47 invasive BCs were found through dynamic contrast between 3 Tesla MR images before surgery and gene expression data

Radiomics on BC were done by dynamic contrast MRI before surgery

Gene expression data were obtained by sequencing whole RNA

To recognize the relationship between 38 previously confirmed image-based features and 186 pathway genes by analysis of gene set enhancement

–

The high radiomics feature related to immune pathways. The use of MRI increases recognition of the tumors with more active immunologically

19

Saha et al. [25]

USA

To investigate the relationships of radiogenomics (BCMSTs with phenotype of MRI)

Cohort

The data of 461 patients to forecast the following:

genomic, molecular, and proliferation characteristics:

estrogen receptor,

tumor surrogate molecular subtype, progesterone receptor, and Ki67

Extracted 529 tumors and surrounding tissue features performed

To predict models of genomic, molecular, proliferative properties, and machine-based learning: Estrogen receptor (ER), progesterone receptor (PR), tumor suppressor, Ki-67, human epidermis, and status of growth factor

AUC:

Luminal A = 0.697

Triple-negative BC = 0.654

ER = 0.649

PR = 0.622

A moderate relationship showed between evaluated imaging features and tumor molecular biomarkers

21

Shin et al. [26]

Korea

To examine radiogenomics (calcification profiles of gene expression in BC patients with mammographic image)

Cohort

168 breast cancer patients were examined base on gene expression analysis

According to the malignancy, mammography calcification is categorized into three groups

RNA was altered to double-strand cDNA. The cDNA was spliced by APE 1 and UDG

The oligonucleotide arrays were used at 45 °C and 60 rpm for 16 h, so spliced and end-labeled cDNAs were hybridized

Use of the multi-array average algorithm to present normalization

–

Compared with cancers without suspected calcifications, immune, inflammatory responses, and defense in BCs with extremely suspected calcifications were reduced in gene analysis

The calcification in BCs reduced the immune activity and is related to ERBB2 mRNA expression in high levels

21

Lin et al. [27]

China

To evaluate the effect of the alterations pathways in BC on radiogenomics

Cohort

88

biopsy-proven breast cancer cases

Radiomics features were taken out from dynamic contrast-enhanced MRI (DCE-MRI) after biopsy of BC cases

By using least absolute shrinkage and selection operator (LASSO) regression analysis and hypothesis testing, modifications' genetic in Myc, cell cycle, p53, receptor tyrosine kinases/RAS (RTK/RAS), and phosphoinositide 3-kinase (PI3K) signaling pathways were forecast by radiomics signatures

To predict the models' power, the receiver operating characteristic curve of area under the curve (AUC) was examined

AUC:

cell cycle = 0.933 Mye = 0.926

PI3K = 0.956

RTK/RAS = 0.940

p53 = 0.886,

Excellent performance in forecasting genetic pathways changes exhibited in radiogenomics analysis of MRI

These changes offer a way to obtain noninvasively tumors' genetic-level molecular features

21

Park et al. [28]

Korea

To forecast MSTs of invasive BC and prognostic biomarkers by using tomography of low-dose perfusion

Prospective

246 patients who were to be treated for aggressive breast cancer

To forecast tumor grade and size, Ki67, HER2, subtypes of molecular, status of lymph node, and HR, 18 CT parameters of cancer were evaluated by performing five machine learning methods

For accuracy, AUC performed

The MSTs of BC are split into four categorizations: triple-negative cancer, HER2 over expression, luminal A, and luminal B

The accuracy of random forest model was higher by approximately 13% on average than accuracy of logistic regression model. The random forest model was found to earn better returns by approximately 0.17 on average of AUC's differences than the logistic regression model

For forecasting molecular subgroups and prognostic biomarkers of BC, radiogenomic machine learning methods performing tomography of low-dose perfusion are a beneficial tool

22

Castaldo et al. [29]

USA

To evaluate radiogenomics (Cancer Genome Atlas in the miRNA expression of invasive carcinomas breast with DCE-MRI)

Pilot

91 invasive breast cancers MRI of T1-weighted dynamic contrast enhancement

The relationship of Cancer Genome Atlas to the miRNA expression of invasive carcinoma breasts was studied with radiomic features of the Cancer Imaging Archive from DCE-MRI

AUC:

ER + versus ER- = 86%

PR + versus PR- = 93%

HER2 + /HER2 = 91%, triple-negative = 91%

Radiological features make it possible to discriminate between the important MSTs of BC and may provide a potential imaging biomarker for advanced medical precision

19

Sutton et al. [30]

USA

To examine the relationship between a texture-based, gene-expression-based, Oncotype DX recurrence score (RS) with MRI features

Retrospective

95 patients with IDC

BC patients were recognized in three ways: 1) DX-RS antibody results 2) PR + , HER2 − , and ER + invasive duct carcinoma 3) before surgery MRI of breast

Based on tissue properties calculated from encapsulated tumors on pre- and post-contrast MR images, features such as grayscale, morphological, and histogram were included

–

A significant relationship among all models that are used with DX-RS antibody image‐based features could forecast the recurrence and benefit of chemotherapy

20

Juan et al. [31]

China

To examine the relationship between the features of DCE-MRI and the expression of Ki-67 in patients who have BC

Retrospective

106 cases with high-Ki-67 expression and 53 cases with low Ki-67 expression

Performing a systematic approach which emphasizes more on extraction of radiomics features and automatic separation of lesion

4 Gy-scale histograms, 5 morphology, and 6 tissue characteristics were found for each lesion

To evaluate the alterations between low and high expressions of Ki 67 and characteristics, statistical analysis was implemented

–

A significant difference showed between gray scale histogram, morphology, and texture characteristics with low Ki-67 expression

20

Mazurowski et al. [32]

USA

To investigate relations between semiautomatically extracted MRI imaging features and BCMSTs

Retrospective

Imaging and genomic data for 48 patients with breast cancer from the Cancer Genome Atlas and the Cancer Imaging Archive

23 features of lesion imaging were extracted from MR images

Morphological, textural, and dynamic properties were extracted

Molecular subgroups were determined based on genomic analysis

The relationship between imaging characteristics and molecular subgroups was evaluated using logistic regression tests and probability ratios

–

Correlation of parenchymal background and tumor enhancement dynamics was found through studying the relationship between MRI features and BC luminal B subtype

20

Woodard et al. [33]

USA

To examine the relationship between the Breast Imaging-Reporting and Data System (BI-RADS) mammography Data System, Breast Imaging Report, MRI features, and risk of BC in patients with positive ER

Retrospective

408 patients identified with invasive breast cancer

Patients with invasive BC were diagnosed who underwent Oncotype DX. Mammographic and MRI features were retrospectively evaluated

• By using features of MRI, the relationship between test relapse score (OD-RS), imaging features, and post hoc pairwise in comparison with OD-RS in line with BI-RADS was evaluated by linear regression

–

Recurrence risk of BC imaging biomarkers is performed by feature of BI-RADS: MRI non-mass enhancement, morphology calcification, mammography and MRI mass margins, and breast density mammography

21

Wan et al. [34]

USA

To recognize features of imaging that took out by computer for breast cancer estrogen receptor (BCER) on DCE-MRI

Retrospective

96 ER-positive breast lesions with high (> 30, N = 41), low (< 18, N = 55), and Oncotype DX recurrence scores

Scores of low and high recurrence in positive ER of breast lesion were used

5 dynamic histogram, 4 intensity kinetics, 4 amplification kinetics, 6 dynamic local binary patterns, 148 tissue kinetics, 3 pharmacokinetics, and 6 shape characteristics are characteristics of each lesion

Linear discriminant analysis classifier was used to evaluate the extracted features by detecting risk of low and high Oncotype DX

Area under the receiver operator characteristic curve helps to examine the performance of categorization

The DLBP and DHoG and attained Az values of 0.80 and 0.84, respectively

A relationship was found between Oncotype DX high- and low-risk stratification with dynamic histogram of oriented gradient (DHoG) and dynamic local binary pattern (DLBP)

For positive ER cancers, computer-derived tissue characteristics of DCE-MRI are increasingly associated with Oncotype DX low- and high-risk classes

19

Yamamoto et al. [35]

USA

To evaluate the radiogenomic preliminary mapping of MR imaging phenotypes to global gene expression patterns in BC

Cohort

353 patients with breast cancer

Gene expression analysis in BC patients was evaluated

On subgroups of 10 patients who experienced MRI, radiogenomic analysis was implemented

To find an association between the imaging data and expression

To find gene sets of interest specific and MRI phenotypes relationships

–

A significant correlation was found between heterogeneous reinforcement patterns of the interferon BC subgroup

To have better understanding of the BC's molecular biology, BC analysis of radiogenomic and MRI is a new method

18

Yamamoto et al. [3]

Korea

To investigate the associations of multi-scale between DCE–MRI phenotypes, long noncoding RNA (lncRNA), and metastasis

Retrospective

47 computational quantitative features extracted from DCE-MR imaging data in a training set (n = 19) to screen for MR imaging biomarkers indicative of poor metastasis-free survival (MFS)

To present biomarkers of poor metastasis-free survival (MFS) of MRI, 47 features were taken out from DCE-MR data of 19 sets of training

By using RNA sequencing—specific of negative binomial distribution alteration expression analysis, the lncRNA molecular feature was clarified

To permit forecast of lncRNA expression and MFS by using polymerase chain response analysis, radiogenomic biomarker was used in the 42 sets of validation

–

A relationship was found between the score of increasing rim fraction of lncRNA radiogenomic biomarker in the DCE-MRI with the predictor of metastatic progression HOX transcript antisense (HOTAIR) expression and early metastasis

20

Tamez-Peña JG et al. [36]

To examine the association of tumor phenotype of digital mammography imaging with BC gene expression signatures

Prospective

71 patients with breast cancer that underwent pre-treatment digital mammograms and tumor biopsies

BC patients who underwent digital mammography before treatment and tumor biopsy

From tumor biopsy, recurrence scores of Oncotype DX and PAM50 were estimated using gene expression microarrays

Multivariate analysis with rigorous validation was used to train recurrence score prediction models

–

Visible phenotype of radiographic was found for BC subtypes and molecular-based relapse risk

19