Molecular imaging in breast cancer: A practical point of view

Abstract

Molecular imaging, defined by positron emission tomography/computed tomography (PET/CT) with ¹⁸F-Fluorodesoxiglucose [¹⁸F]FDG, plays a pivotal role in the clinical management of breast cancer (BS), providing essential information for accurate staging, early detection of recurrence and supporting the selection and monitoring of personalized treatment strategies. Magnetic resonance imaging (MRI), the most sensitive technique in BC detection, offers a potentially valuable support, as a partner of PET, in hybrid PET/MRI scans, especially in locoregional disease assessment or some distant metastases locations. In addition, radiomics-derived parameters and machine learning approaches can refine molecular stratification with insights into treatment and prognosis.

Introduction

Breast cancer (BC) remains the most prevalent malignancy among women in Spain, with approximately 37,682 new cases projected for 2025, according to national cancer registries. On a global scale, BC ranks as the second most commonly diagnosed cancer and the fourth leading cause of cancer-related mortality, highlighting its significant burden on healthcare systems and the urgent need for continued innovation in oncologic care [1]. The integration of advanced imaging modalities into routine clinical practice has significantly enhanced diagnostic precision and therapeutic planning, contributing to improved patient outcomes. Hybrid imaging, such as positron emission tomography/computed tomography (PET/CT) or positron emission tomography/magnetic resonance imaging (PET/MRI) with ¹⁸F-Fluorodesoxiglucose [¹⁸F]FDG, allows the integration of metabolic, anatomical and functional information (cell density and perfusion data) in the same examination. [¹⁸F]FDG PET/CT provides essential information for accurate staging, facilitates early detection of metastatic disease, and supports the selection and monitoring of personalized treatment strategies. The recently published European Association of Nuclear Medicine/Society of Nuclear Medicine and Molecular Imaging (EANM/SNMMI) guideline, endorsed by major BC societies as American College of Radiology (ACR), European Society of Surgical Oncology (ESSO), European Society for Radiotherapy and Oncology (ESTRO), European Society of Breast Imaging/European Society of Radiology (EUSOBI/ESR), and European Society of Breast Cancer Specialists (EUSOMA), underscores the clinical relevance of [¹⁸F]FDG-PET/CT in managing no special type (NST) BC [2].

This review compares recent recommendations with those from the National Comprehensive Cancer Network (NCCN), the European Society for Medical Oncology (ESMO) and ABC (Advanced Breast Cancer), highlighting key overlaps and innovations [3].

In the following sections, the aim of [¹⁸F]FDG-PET/CT and PET/MRI is described, under a practical perspective of use of both diagnostic techniques, for initial staging, response evaluation, recurrence detection and their novel applications of artificial intelligence and radiomics.

Initial staging

For adequate management of the disease, accurate locoregional and distant staging is essential, as it determines the type of surgery, the indication for systemic therapy, the fields of radiotherapy irradiation and moreover provides prognostic information. MRI is the most sensitive technique in BC detection and offers a valuable support in the locoregional and distant disease evaluation. The synergy and integration of information allowed by PET/MRI offer advantages apart from reducing radiation use by avoiding the need for a CT scan, which typically provides no additional information in these patients.

Attending to the classification system of the American Joint Committee on Cancer (AJCC), the aim of [¹⁸F]FDG-PET/CT and PET/MRI is described for T (primary tumor), N (nodal status) and M (distant metastases) staging in the next sections [4].

T stage

[¹⁸F]FDG-PET/CT has a sensitivity and specificity in BC detection, ranging from 48% to 96% and 73% to 100%, respectively [5,6]. The sensitivity is very limited for lesions less than 6 mm, attributed to the partial volume effect, explained by the reduced spatial resolution and respiratory movement interference. In addition, tumor characteristics such as proliferation rate, tumor grade, histological subtype and molecular profile may affect lesion detection capacity on PET/CT. For example, invasive lobular carcinoma (ILC), low Ki-67 proliferation, grade 1 or 2 or well-differentiated hormone receptor (HR) positive tumors tend to display lower [¹⁸F]FDG activity, potentially leading to false negative [¹⁸F]FDG-PET/CT scans [7,8].

Up to now, breast MRI is the most sensitive modality (90–99%) for detecting BC, assessing local tumor spread and allowing the identification of occult breast lesions. However, its specificity is moderate (72–89%), so false-positive findings require additional imaging or biopsies. The combined PET/MRI has been shown to increase accuracy [9,10]. A study assessing 49 patients and 83 tumor lesions was demonstrated that there was no significant difference between T-staging by MRI or [¹⁸F]FDG-PET/MRI, with [¹⁸F]FDG-PET/CT being less accurate (82% vs. 68%) [11]. Taneja et al. [12] in a pilot study demonstrated greater diagnostic confidence in the joint T assessment of selective breast [¹⁸F]FDG-PET/MRI compared to MRI or [¹⁸F]FDG-PET/CT separately, resulting in the detection of satellite lesions that cause multifocality and multicentricity in 58% of patients and of contralateral synchronous lesions in 5.5%. A recent meta-analysis reported a sensitivity of 95%, specificity of 91%, and area under the curve (AUC) of 0.96 in the detection of tumor lesions by [¹⁸F]FDG-PET/MRI [13]. In addition, the selective acquisition of breasts on prone position is superior compared to-whole body imaging in the supine position, with the correct identification of BC in 97% and 87%, respectively [14,15].

Thus, although [¹⁸F]FDG-PET/MRI is not currently recommended for BC diagnosis, its utilization could improve the diagnostic accuracy of MRI, enabling a more comprehensive diagnostic strategy (Fig. 1).

Figure 1.

Case A. Selective breast 18F-FDG PET/MRI: axial PET (a), axial T2-weighted MRI (b) and axial fused PET/MRI (c) focused on the right breast showing multifocal lesions. Solid and spiculated hypermetabolic mass (44 × 35 mm) in the lower outer quadrant of the right breast with a SUVmax of 8.8. It is associated with a 17 mm hypermetabolic nodule (SUVmax 2.9) at the junction of the lower quadrants, approximately 27 mm deep from the tumor mass. Case B. Selective breast 18F-FDG PET/MRI: axial PET (a, d), axial T2-weighted MRI (b, e) and axial fused PET/MRI (c, f) (upper line focused on the right breast lesion (arrow), lower line focused on the left breast lesion (arrow). Hypermetabolic lesion at the junction of the outer quadrants of the right breast, posterior third, with spiculated margins of 19 mm (SUVmax 7.6). A solid hypermetabolic nodule measuring 9 mm (SUVmax 2.3) in the left breast at the depth of the junction of the upper quadrants, located prepectorally.

N stage

Breast cancer usually disseminates loco-regionally to the ipsilateral axillary lymph nodes (ALNs) but can also spread loco-regionally to ipsilateral internal mammary or supraclavicular lymph nodes up to stage N3c [4,16].

The diagnostic performance of [¹⁸F]FDG-PET/CT is inferior to sentinel lymph node biopsy (SLNB) for nodal staging in BC. Thus, in early-stage tumors, PET cannot replace SLNB, owing to its low sensitivity for micrometastases. However, the high specificity of PET, in a patient with suspected advanced disease, might justify avoiding the need for SLNB [17,18]. Moreover, in this group, [¹⁸F]FDG-PET/CT can lead to upstaging in up to 24% of patients (mainly due to the identification of N3 disease) and downstaging in 16% of patients [19–21]. In addition, [¹⁸F]FDG-PET/CT has a high positive predictive value in diagnosing internal mammary lymph nodes [22].

The percentage of patients with extra-ALN involvement detected by [¹⁸F]FDG-PET/CT in locally advanced BC ranges from 10% to 29% [22,23]. Thus, in this setting, given the high prevalence of lymph node involvement (up to 80%), PET/CT has a high negative predictive value in lymph nodes, adding to its usefulness [23,24].

[¹⁸F]FDG-PET/MRI offers a better evaluation of the axillary region in comparison with conventional tests, increasing the diagnostic confidence for nodal involvement based on the combined [¹⁸F]FDG uptake on PET and diffusion restriction on MRI. A meta-analysis including 4 studies showed a pooled sensitivity, specificity and AUC of 94%, 90% and 0.96, respectively [9,12,25] (Fig. 2). Morawitz et al. [26] demonstrated a higher number of regionally infiltrated lymph nodes on [¹⁸F]FDG-PET/MRI compared to MRI and CT (41.2% vs. 30.8% vs. 20.8%). On the other hand, Taneja et al. [12] showed a sensitivity of 60% and 93.3% on PET and MRI, respectively, for the detection of ALN. Therefore, some authors propose clinical trials evaluating the reliability of [¹⁸F]FDG-PET/MRI to replace SLNB in axillary staging, currently considered as the gold standard [27].

Figure 2.

Examples of lymph node involvement assessed by 18F-FDG PET/MRI on axial PET (a), axial T2-weighted MRI (b) and axial fused PET/MRI (c). A. right axillar lymph node (SUVmax 2). B. Right retropectoral hypermetabolic lymph node (SUVmax 4). C. Left supraclavicular hypermetabolic lymph node (SUVmax 4). D. Right internal mammary hypermetabolic lymph node (SUVmax 3).

M stage

Dissemination to contralateral axillary, internal mammary, or supraclavicular lymph nodes or to ipsilateral or contralateral cervical lymph nodes is regarded as distant metastasis (or stage M1). The diagnosis of distant metastases in the initial staging phase of BC is significant because it shifts clinical management from a curative to a disease stabilization approach.

Breast cancer can potentially spread to any organ, but the most common sites are the skeleton, liver, lung, and brain. Bone metastasis can be lytic, sclerotic, mixed, or intramedullary, without obvious bone changes. [¹⁸F]FDG-PET/CT is better than bone scanning (BS) in identifying lytic and intramedullary metastases, although [¹⁸F]FDG-PET/CT is less efficient in identifying sclerotic bone metastases. However, these non-avid lesions are often identified in the CT component of PET scans [28,29]. Regarding the aim of different imaging techniques on bone metastases detection, a recent metanalysis reported that [¹⁸F]FDG-PET/CT and MRI outperformed fewer sensitive modalities as CT and BS with values of sensitivity and specificity of 94% and 98% for [¹⁸F]FDG-PET/CT, 95% and 93% for ¹⁸F-sodium fluoride [¹⁸F]-NaF-PET/CT, 94% and 93% for MRI, 70% and 98% for CT and 83% and 96% for BS, respectively [30].

Different from invasive ductal carcinoma (IDC), metastatic invasive lobular carcinoma (ILC) more commonly features sclerotic bone metastases without [¹⁸F]FDG uptake and metastases to the gastrointestinal tract and serosa. Thus, [¹⁸F]FDG-PET/CT is less likely to reveal unsuspected distant metastases in patients with ILC compared to IDC [31]. In addition, each molecular subtype has different patterns of metastatic spread [32–35]. For example, bone metastases are more common in HR (+) tumors, whereas visceral metastases occur more often in HR (−) as triple-negative breast cancers (TNBC) and human epidermal growth factor receptor 2 (HER-2) enriched subtypes [33,36–38].

Patients younger than 40 years usually have more aggressive phenotypes of BC, as TNBC, associated with distant metastases at an earlier clinical stage [39]. In addition, inflammatory tumors, which are more easily detected by [¹⁸F]FDG-PET/CT, have a fast spread [40,41]. Thus, [¹⁸F]FDG-PET/CT can be useful in these higher-risk disease patterns [39–41].

Baseline [¹⁸F]FDG-PET/CT enables overall upstaging, when compared with conventional imaging techniques (CITs), with percentages depending on the previous clinical stage: stage IIA (13%), stage IIB (19%), IIIA (34%), IIIB (41%), and IIIC (35%) [19–21]. The identification of distant metastases on [¹⁸F]FDG-PET/CT is especially relevant in locally advanced BC (up to 26%), being superior to BS and CT of the chest/abdomen/pelvis and leading to changes in therapy decision in up to 18% [20,42–45].

Regarding the use of [¹⁸F]FDG-PET/MRI, PET adds sensitivity to MRI in millimeter lesions. In this regard, Garcia et al. [46] in a prospective study reported internal mammary chain infiltration in 34% of patients with stage ≥ IIB, allowing tailored therapy, with thoracic radiotherapy implementation in 24% of patients.

Previous reports define that [¹⁸F]FDG-PET/MRI led in change in treatment management in up to 33% and 12%, compared with CIT and contrast-enhanced [¹⁸F]FDG-PET/CT, respectively [12,47–49]. Moreover, although in general terms, [¹⁸F]FDG-PET/MRI seems comparable to PET/CT in extra-ALN detection, the former is more accurate than conventional imaging techniques (CITs), even PET/CT, in bone and liver metastases detection [50,51]. A recent meta-analysis including 3 studies of BC (182 patients) reported a higher sensitivity and specificity of [¹⁸F]FDG-PET/MRI with respect to [¹⁸F]FDG-PET/CT; 95% versus 87% and 96% versus 94%, respectively [52]. In a study by Catalano et al. [53], the combined use of whole-body diffusion-weighted imaging (WB-DWI), WB-PET/CT and WB-PET/MRI correctly staged IDC in 64% of patients. In cases of discordance, WB-PET/MRI had superior accuracy (98%) with respect to WB-PET/CT (75%) and WB-DWI (84%). The improved staging performance of WB-PET/MR was likely due to the higher sensitivity of MRI in detecting non-FDG avid lesions (permeative bony and sub-centimeter hepatic metastases). Thus, if available, [¹⁸F]FDG-PET/MRI would be considered for staging patients with locally advanced IDC or high-risk cases, as it shares the same clinical indications as [¹⁸F]FDG-PET/CT [54] (Fig. 3).

Figure 3.

Whole-body 18F-FDG PET/MRI: axial PET (a), axial T2-weighted (b), axial DWI axial (c) and axial fused PET/MRI (d). Distant metastases localized on: A. right iliac bone (SUVmax 3.1); B. subcapsular liver lesion on segment VII, measuring 10 mm (SUVmax 3.7); C. right cerebellar intra-axial lesion; and D. micronodule in the upper lobe of the right lung (7 mm, SUVmax 3).

Treatment response

Neoadjuvant treatment

Early response

For early evaluation, after the first or second cycle of neoadjuvant chemotherapy (NAC), [¹⁸F]FDG-PET/CT has showed good sensitivity (80–85%), but low specificity, to predict early histopathological response to NAC in the primary tumor, with independence of BC molecular subtypes [55–58]. In addition, comparing PET to MRI to predict pathologic complete response (pCR), PET is more sensitive and MRI is more specific [59]. Thus, [¹⁸F]FDG-PET/MRI would offer a superior one-stop shop technique, being useful in the non-invasive prediction of pCR, with higher accuracy compared to PERCIST [60].

It is worth mentioning that the response rate to NAC varies among the different molecular subtypes. Thus, tumor subtype and the treatment used should be considered to assess response more effectively. HR (+) tumors, less sensitive to chemotherapy and with less avidity for [¹⁸F]FDG, are not suitable for response assessment with PET/CT [61,62]. However, for TNBC, more chemo-sensitive, PET/CT response is better correlated with pCR at surgery, whereas in HER (+) BC, PET/CT seems useful in modulating NAC according to the early metabolic response [63–66].

Unfortunately, the optimal cut-off of the maximum standardized uptake value (SUVmax) to predict response varies between the studies that limit its final applicability. Therefore, validation of [¹⁸F]FDG-PET/CT as an early response marker probably will be possible in these more molecular aggressive tumors in some ongoing clinical trials (NCT04882371, NCT05710328, NCT01396655, NCT03161353).

End of therapy response

[¹⁸F]FDG-PET/CT is not very sensitive at the end of treatment to reveal the residual primary tumor tissue [67–69]. Thus, the rate of false negatives is not negligible, although the false positives are very limited. Regarding the lymph node response, the limitations are greater, and these are not only for PET but also for MRI and axillary ultrasound [70,71]. Therefore, based on there being no accurate non-invasive technique to identify patients with a complete axillary response after NAC, SLNB, with or without the removal of marked initially positive lymph nodes, should be considered in patients with an imaging-based negative axilla.

On the other hand, a metabolic response on [¹⁸F]FDG-PET/CT is associated with prognosis, with the results related to disease recurrence and survival in BC patients [69].

Thus, [¹⁸F]FDG-PET/CT seems to perform better in the early prediction of response and MRI at the end of NAC. Although the experience is limited, using prone breast-selective [¹⁸F]FDG-PET/MRI, correlations between [¹⁸F]FDG-PET and MRI biomarkers have been described [72]. However, a recent consensus defined that since there is no optimal and reproducible cut-off value of SUV, suitable for different centers, each institution should select the imaging technique based on its availability and diagnostic experience of the imaging team [73] (Figs. 4 and 5).

Figure 4.

Basal of [¹⁸F]FDG-PET/CT of a 46-year-old woman diagnosed with CDI of the right breast pure HER2(+) phenotype with lymph node involvement. (A) also detects adenopathic involvement in the right supraclavicular region and bone involvement in the sacrum and right femur of [¹⁸F]FDG-PET/CT after two cycles (B) shows partial response in breast tumor and lymph nodes and complete metabolic response at the end of treatment (C). Mastectomy and lymphadenectomy revealed a complete histological response in both locations. [¹⁸F]FDG-PET/CT at the end of treatment correctly cataloged both responses.

Figure 5.

Basal [¹⁸F]FDG-PET/CT of a 48-year-old woman diagnosed with CDI of the left breast, basal phenotype with lymph node involvement. Images in axial slices showing the left breast lesion with axillary lymph node involvement (A) with no response in [¹⁸F]FDG-PET/CT after two cycles of neoadjuvant treatment (B), partial response at the end of treatment (C) and complete metabolic response after second line of chemotherapy with the addition of taxanes (D). After quadrantectomy and lymphadenectomy, a complete histological tumor and lymph node response was determined. [¹⁸F]FDG-PET/CT correctly cataloged the evolution of responses to neoadjuvant therapy.

Distant metastases response

The prognosis for women diagnosed with BC is generally favorable, with a 5-year overall survival (OS) rate of 90%. Once metastasis occurs, prognosis worsens significantly, with a 5-year survival rate of only 26% [74]. Thus, as therapeutic options increase, accurate assessment of treatment response becomes critical, particularly in distinguishing oligometastatic from widespread disease and in detecting progression [75].

Response evaluation in metastatic breast cancer (MBC) relies on CIT, such as contrast-enhanced CT (CE-CT) and BS [3]. However, CE-CT has low sensitivity for bone and low specificity for liver metastases [76]. Moreover, approximately 80% of BCs are luminal subtypes, which preferentially metastasize to bone. This biological behavior pushes challenges for early detection and treatment monitoring using CIT. BS lacks specificity and has only 51% sensitivity for osteolytic lesions, while CT requires 30–50% bone mineral loss to detect metastases. In addition, BS has lower spatial resolution with respect to PET/CT and its interpretation is complex due to density changes, interference from bisphosphonates or anti-RANKL agents and flare-up phenomenon, commonly seen the first 6 months following the introduction of hormonal drugs or chemotherapy [77,78]. Like ^99mTc-methylene diphosphonate for BS, [¹⁸F]NaF (as a PET tracer) accumulates in areas of active bone remodeling and has shown superior sensitivity compared to BS and CT [79,80]. The quantification of [¹⁸F]NaF uptake in bone metastases has been utilized in several studies to assess bone response, but similar to BS, [¹⁸F]NaF-PET/CT is also susceptible to the flare phenomenon [81].

[¹⁸F]FDG-PET/CT can assess response immediately after one or two cycles and detect progression earlier than CE-CT [76,82]. However, [¹⁸F]FDG-PET/CT may be less effective in ILC, which often metastasizes to the gastrointestinal tract, peritoneum, and bone, and typically shows low Ki-67 and [¹⁸F]FDG uptake [83].

Moreover, the use of PET has prognostic implications. In a cohort of 300 MBC patients, 5-year survival exceeded 40% with [¹⁸F]FDG-PET/CT monitoring, versus 16% with CT alone [84]. In addition, PET Response Criteria in Solid Tumors (PERCISTs) (one-lesion approach) has demonstrated being a superior predictor of progression-free and disease-specific survival compared to response evaluation criteria in solid tumors (RECISTs) [85].

Based on the superiority of [¹⁸F]FDG-PET/CT with respect to CT in evaluating bone metastases, Joint EANM-SNMMI guidelines support the use of [¹⁸F]FDG-PET/CT [3,86]. However, this technique lacks standardized and validated criteria for widespread clinical use [74]. NCCN endorses RECIST or World Health Organization (WHO) criteria, EANM-SNMMI recommends PERCIST or European Organization for Research and Treatment of Cancer (EORTC) criteria. In addition, in immunotherapy, EANM guidelines should be followed [3].

The use of [¹⁸F]FDG-PET/MRI, limited in these contexts, seems more useful in liver metastases response based on the limitations of [¹⁸F]FDG-PET/CT [87].

Recurrence detection

Following primary treatment, recurrence may occur in up to 35% of patients within ten years of the initial surgery, with an incidence of distant metastasis in up to 70% [88].

Early detection and accurate restaging of recurrence is the main goal during the surveillance of patients; mainly, it is important for guiding appropriate treatment, and the survival rate is likely to improve if early detection is established [75]. In general, systemic therapy is used in all patients with recurrent/metastatic disease; however, isolated locoregional disease or single sites of metastatic recurrence can also be treated with surgery and radiation therapy with a curative intent [89].

Four meta-analyses have showed the high performance of [¹⁸F]FDG-PET/CT to detect recurrent BC [90–93]. [¹⁸F]FDG-PET/CT imaging has demonstrated high sensitivity in detecting BC with a pooled sensitivity, specificity and accuracy of 90%, 81% and 93%, respectively. In addition, compared with CIT, [¹⁸F]FDG-PET/CT offers a whole-body approach to determine the correct extent of disease [94]. [¹⁸F]FDG-PET/CT improves prognostic stratification by distinguishing patients with isolated loco-regional recurrence from those with distant metastases, defining clearly the total burden of disease that has to be treated. This has a direct impact on the management in different scenarios (Fig. 6).

Figure 6.

(A) Local breast recurrence. (B) local recurrence and axillary involvement. (C) Oligorecurrence. (D) Metastatic disease.

According to clinical guidelines [3], the follow-up of an asymptomatic patient with early-stage BC treated with curative intent is based on regular physical exams and annual breast imaging, such as mammography. In the absence of clinical signs or symptoms suggestive of recurrence, there is no indication for laboratory or imaging studies for metastases screening. Joint EANM-SNMMI guidelines agree with this principle, stating that [¹⁸F]FDG-PET/CT is useful in recurrence suspected by clinical examination, CIT (even equivocal), or tumor marker elevation as cancer antigen (CA) 15.3 or carcinoembryonic antigen (CEA) [2]. When tumor markers have a median range of 58.9 U/ml for CA 15.3 and of 12.4 U/ml for CEA, [¹⁸F]FDG-PET/CT increases sensitivity, specificity, positive predictive value and negative predictive value to 93.6%, 85.4%,96.7% and 74.5%, respectively [95]. On the other hand, in the case of clinical suspicion of relapse, [¹⁸F]FDG-PET/CT can reveal recurrence, even with negative tumor markers [93].

[¹⁸F]FDG-PET/CT is effective in detecting distant metastases, and can also substitute for CE-CT and/or BS in the detection of bone metastases, and also in showing locoregional recurrence, especially in the chest wall and axillary and extra-axillary lymph node regions, being more effective than CT or MRI in detecting lymph node recurrence [96].

[¹⁸F]FDG-PET/CT can enhance the identification of isolated metastatic lesions, mainly in the bone, that are missed by the CIT (CT or BS). If there is suspicion of oligometastatic disease, clinical guidelines recommend whole body staging, preferably with [¹⁸F]FDG-PET/CT [75]. This is important due to changes in recurrent tumor histology, such as the common loss of HR expression which it is crucial and directly influences therapeutic strategy, improving RT planning of those isolated metastases after a curative rescue surgery.

Limitations of [¹⁸F]FDG-PET/CT are its poor sensitivity in detecting brain metastases, due to the high physiological uptake of [¹⁸F]FDG, sclerotic bone metastasis based on their reduced metabolism and subcentimetric lung nodules. However, some of them can be diagnosed by the CT component [97]. For brain metastases, MRI is the preferred imaging test. However, in those patients with distant metastases, [¹⁸F]FDG-PET/CT can serve as a baseline study to assess response to chemotherapy.

[¹⁸F]FDG-PET/MRI is useful for re-evaluation, showing superior sensitivity and accuracy over CT, MRI, and PET/CT. The meta-analysis by Lin et al. [50] showed an overall sensitivity and specificity of 98% and 87% in the patient-specific analysis and 91% and 95% in the lesion-specific analysis. [¹⁸F]FDG-PET/MRI has demonstrated a better detection rate than MRI in cases of local recurrence. In addition, in oligometastatic patient candidates for treatment with stereotactic body radiotherapy, PET/MRI is of great utility for the planning of treatment and subsequent follow-up [9,98]. De Mooii et al. [99], in a systematic review, demonstrated greater sensitivity in detecting metastatic lesions in patients with BC with [¹⁸F]FDG-PET/MRI compared to [¹⁸F]FDG-PET/CT (89% vs. 77%), differentiating between liver metastases (80–100% vs. 70–75%) and bone metastases (92–98% vs. 69–99%).

Machine learning and radiomics

Machine learning (ML) and deep learning, branches of artificial intelligence (AI), aim to analyze, classify, and judge huge digital information to help formulate algorithms for diagnosis and therapy [100,101]. In addition, these approaches are especially useful to manage medical imaging information, basically because a great of semiquantitative variables can be obtained.

Medical imaging enables the full-scale mapping of intratumoral or peritumoral areas via a non-invasive approach, acting as a mediator, bridging the gap between the phenotype and genotype of tumors. Radiomics refers to the process of extracting and transforming images into quantitative features to gain insights into tumor characteristics that can reveal biological information and prognosis about diseases. Radiomic workflow involves several steps that go from image acquisition and segmentation to feature extraction and selection, until model construction. Imaging features can describe things like the shape, size, texture, and intensity variations within the image.

Radiomics and ML are inseparable partners in molecular imaging, because the latter defines the application of radiomics in the medical field. In BC metabolic imaging, these applications mainly are focused on disease detection vs. molecular profile prediction, treatment response and prognosis.

Tumor detection and molecular classification

Promising results, using [¹⁸F]FDG-PET/CT, have been obtained in the characterization of the primary tumor characteristics, in particular, the molecular subtypes, although these have not been confirmed by all the studies analyzed [102]. SUV-derived parameters have higher values in more aggressive molecular phenotypes. Regarding [¹⁸F]FDG-PET/MRI, the experience is more limited although the AI model seems allowing for highly accurate differentiation of benign vs. malignant lesions [103].

Several papers, focused on the potential diagnostic role of PET radiomics for predicting ALN metastases, have described significant accuracy using predictive models with ML, using [¹⁸F]FDG-PET/CT and [¹⁸F]FDG-PET/MRI [104–107].

Treatment response

Accurate prediction of NAC response is critical for tailoring chemotherapy regimens or determining an appropriate surgery strategy [108,109].

The possibility of predicting BC pathological response to NAC based on the baseline [¹⁸F]FDG-PET/CT, without the need for an interim study, is a focus of recent discussion. In a recent review, eight out of 13 analyzed studies indicated an association between [¹⁸F]FDG-PET-based tumor uptake heterogeneity features and prediction of response to NAC [110]. Using [¹⁸F]FDG-PET/MRI, radiomic analysis at baseline, allowing for the prediction of complete pathological response, especially in HR(+)/HER2(−) patients [111].

However, features associated with predicting response to NAC varied between studies, being necessary, reproducible and applicable findings.

Prognosis

High [¹⁸F]FDG uptake, using metrics such as the SUVmax and volumetric variables such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG), correlates with tumor aggressiveness and is associated with a worse prognosis [112–114].

Pak et al. [114] concluded that patients with a high MTV and TLG from the primary tumor have a higher risk of adverse events and patients with a high TLG from whole-body tumor burden have a higher risk of death, therefore suggesting that these volumetric parameters should be routinely used when reporting scans. However, no specific cut-off values for these metrics can be recommended currently, as these values differ widely among the data published [115].

Summarizing, based on BC is a heterogeneous disease with a significant chance that some of its features, including significant characteristics, remain undetected, metabolic imaging offers an integrated and global information with predictive and prognostic value. Such information is especially relevant in the current era of personalized medicine focused on individually tailored treatments. Currently, based on the limited generalizability of radiomic models, some research is focusing on evaluating and optimizing the robustness of these models to improve their applicability.

Informed consent statement

Not applicable.

Ethical considerations

Patient data proceed from a previous approved study.

Funding

This research received no external funding.

Artificial intelligence

AI was not used in the redaction of the present manuscript.

Authors' contributions

AMGV, CMSB, JRGG and MCR collected and reviewed literature, wrote the paper and designed figures; AMGV integrated and reviewed the paper. All the authors read and edited the manuscript.