Magnetic resonance biomarkers in detection and characterisation of ovarian cancer: is magnetic resonance spectroscopy ready for clinical use?

Abstract

Ovarian cancer is a major cause of cancer-related mortality in women, largely due to late-stage diagnosis and challenges in preoperative characterisation of adnexal masses. Multiparametric MRI, incorporating diffusion-weighted (DW) and dynamic contrast-enhanced (DCE) imaging, has improved lesion characterisation and led to the development of the Ovarian-Adnexal Reporting and Data System (O-RADS) MRI for standardised malignancy risk stratification. Despite its demonstrated clinical value, O-RADS MRI relies on qualitative assessment and lacks metabolic information. Magnetic resonance spectroscopy (MRS) provides non-invasive insight into tissue biochemical composition. It offers potential for differentiating benign and malignant ovarian lesions that are indeterminate on ultrasound or MRI. This review discusses current functional MRI biomarkers within the O-RADS framework and evaluates the emerging role of MRS as potential biomarker for ovarian cancer characterisation. Quantitative metrics derived from DWI, including apparent diffusion coefficient (ADC) and histogram analyses, and semi-quantitative and quantitative DCE parameters, have demonstrated utility in differentiating benign, borderline, and malignant lesions. Variability in acquisition protocols, post-processing, and region-of-interest selection, along with motion artifacts and magnetic field inhomogeneities, limits routine clinical implementation. Advanced diffusion models and whole-lesion histogram approaches improve lesion characterization, particularly in indeterminate O-RADS MRI 4 masses, but no single parameter reliably discriminates all lesion types. Early MRS studies identified metabolites such as choline (Cho), lactate (Lac), lipids (Lip), and N-acetylaspartate (NAA) as potential malignancy indicators, with Cho/Cr ratio frequently investigated. Diagnostic performance has been inconsistent due to small cohorts, variable metabolite thresholds, and heterogeneity in acquisition and analysis protocols. Technical limitations, including low signal-to-noise ratio, motion artifacts, field inhomogeneities, and voxel mislocalisation, further limit clinical adoption. Standardized MRS acquisition, voxel localization, shimming, denoising, and post-processing are needed. Community-endorsed consensus guidelines, similar to those in brain MRS, could improve reproducibility, comparability, and reliability. Currently, MRS remains investigational but may serve as a complementary biomarker for O-RADS MRI 4 lesions. Harmonised protocols and consensus guidelines are required to support its clinical translation.

Background

Ovarian cancer is one of the most prevalent cancers in women [1] and the 4th most common cause of cancer-related deaths among women, causing more deaths than any other gynaecologic malignancy [2]. Ovarian tumours can be classified as benign, borderline or malignant [3]. The majority of ovarian cancers are detected at late stages (III or IV) with spread outside the pelvis, either due to being asymptomatic in early stages or the lack of effective screening tests [2, 4]. Biopsy is generally not recommended for primary evaluation of an adnexal mass because of the risk of dissemination and sampling error [5]. However, image-guided biopsy is routinely performed in patients with suspected advanced disease who are planned for neoadjuvant chemotherapy to confirm diagnosis and support molecular or genomic characterisation [6]. Consequently, a significant number of women, having benign masses indeterminate in ultrasound, undergo unnecessary or inappropriately aggressive surgery, resulting in increased fertility loss and morbidity [7]. A non-invasive diagnostic technique with high sensitivity and specificity for preoperative characterisation and risk stratification of indeterminate ovarian/ adnexal masses is currently missing in clinical practice.

In ovarian cancer, MRI of the pelvis plays a crucial role in lesion differentiation and characterisation, particularly in indeterminate masses [8, 9], which are often first picked up on ultrasound. Several studies have been performed to assess the diagnostic performance quantitatively and semi-quantitatively of functional imaging using MRI, such as diffusion weighted imaging (DWI) and dynamic contrast enhanced (DCE) MRI [3, 10–13]. DWI has been proved beneficial in lesion characterisation, increasing the diagnostic confidence by approximately 15%, when it is added in the conventional anatomical MRI [14]. DCE-MRI has also increased the specificity of malignancy prediction, with the presence of enhancement indicating an increased likelihood of malignancy [15]. All these highlighted the value of anatomical and functional MRI to accurately characterise malignancy in indeterminate adnexal lesions in ultrasound and introduced the clinical need of a scoring reporting system, determining the framework for the management of patients with ovarian and adnexal lesions.

Therefore, a scoring system for reporting ovarian and adnexal lesions was introduced in 2013 [16], the Ovarian-Adnexal Reporting and Data System (O-RADS) MRI. This system, analogous to other American College of Radiology (ACR) reporting and data systems (RADS), allows risk stratification of the likelihood of malignancy of ovarian masses, aiming to maximise patient management and promote standardisation. This 5-point system evaluates the anatomical high-resolution T1-weighted (T1W) and T2-weighted (T2W), DWI and DCE-MRI. Multiple groups reviewed the framework and the use of O-RADS MRI [9, 17–19]. Rockall et al. [19] highlighted the technical aspects for high diagnostic quality MRI acquisitions, including patient set-up, magnetic field strength, protocol requirements, and strategies to reduce artifacts. Practical guidance for O-RADS MRI use was presented, to support systematic assessment of suspicious anatomical characteristics [18]. While O-RADS MRI provides a standardised reporting framework with high accuracy, this has been confirmed in a recent multi-centre study [7].

Furthermore, O-RADS MRI functional imaging techniques provide information on the cellularity and vascularity of the suspected area of interest, but not its biochemical composition [20, 21]. Consequently, there is still a need to establish imaging biomarkers of malignancy for lesion differentiation, especially at lesions that may have one or more concerning feature on MRI. For these cases, detection of metabolic changes might be valuable and predictive of treatment response. To assess those changes, different groups have investigated magnetic resonance spectroscopy (MRS) as a potential imaging biomarker for early ovarian cancer diagnosis over the last 20 years (Table 1). Despite promising results, its clinical translation is still vague. This article will review and discuss the proposed imaging biomarkers in ovarian cancer using the O-RADS MRI functional imaging metrics, and the utility of MRS.

Table 1.

Characteristics of research studies in magnetic resonance spectroscopy (MRS) in ovarian cancer

Author/ year	Type of tumours	Aim of the study	Number of tumours	Post-processing analysis	Metabolites
Okada 2001	Intrapelvic tumours	Measure metabolites of large tumours	23	Semi-quantitative	Lac, Cho
Cho 2002	Adnexal lesions	Characterise spectral parameters of adnexal lesions	31	Integrated peak	Lip
Stanwell 2008	Ovarian tumours	Characterise ovarian neoplasms	14	Semi-quantitative	Lip, Lac, Cho/Cr
Booth 2009	Gynaecological lesions	Characterise spectra of benign and malignant gynaecological lesions	39/57 analysed	Not mentioned	Cho
Mc Lean 2009	Ovarian masses	Detect and quantify metabolites in the primary and metastatic masses	23	Quantitative	Cho
Esseridou 2011	Ovarian masses	Evaluate the diagnostic performance of tCho as a marker of malignancy	23	Visual assessment	Cho
Takeuchi 2011	Ovarian tumours	Distinguish between mucinous and non-mucinous ovarian tumours	32	Semi-quantitative and quantitative	NAA
Sala 2012	Advanced ovarian cancer	Evaluate the response of platinum-based neoadjuvant chemotherapy	22	Quantitative	Cho
Ma 2015	Adnexal tumours	Differentiate benign from malignant solid adnexal tumours	69	Semi-quantitative	Cho, Cr, NAA, Lip
Malek 2015	Adnexal tumours	Improve diagnostic accuracy in characterization of adnexal masses by detection of choline peak using 1 H MRS in comparison to internal reference at 3.0T	46	Not mentioned	Cho
Mansour 2019	Ovarian masses	Assess the diagnostic performance in the screening of malignancy in the ovarian masses	245	Semi-quantitative	Lac, Lip, NAA, acetone, Cho, Cr
Ma 2019	Epithelial ovarian tumours	Differentiate borderline from malignant epithelial ovarian tumours	82	Semi-quantitative	NAA, Cho, Cr, Lac, Lip
Takeuchi 2021	Ovarian tumours	Distinguish benign from malignant mucinous ovarian tumours	43	Semi-quantitative	NAMC, Lip, tCho
Li 2021	Ovarian neoplasms	assess the additional value for risk stratification of sonographicaly indeterminate ovarian neoplasm	21	Quantitative	Cho, Lip

Lac: lactate, Cho: choline, Lip: lipid, Cr: creatine, NAA: N-acetyl aspartate, NAMC: N-acetyl mucinous compounds, tCho: total choline

O-RADS MR imaging biomarkers

In the last decade, the research community has been focusing on the development of quantitative assessments of functional MRI techniques, aiming to derive an objective relation between the quantitative values and tissue properties, such as cellularity and vascularity [22]. In oncology, these quantitative metrics could serve as imaging biomarkers, allowing the lesion detection and characterisation, the prediction of treatment response, thus improving patient management and care. Research groups have concentrated on the evaluation and standardisation of imaging biomarkers in tumours of different types of cancers, revealing their value and their requirement [23].

In ovarian cancer, using imaging biomarkers, that are both non-invasive and do not use ionising radiation, for indeterminate lesion characterisation, would spare patients with benign or low-risk lesions from having invasive surgery thus increasing fertility via sparing ovaries and reducing morbidity (Table 2). In a systematic review and meta-analysis, Rizzo et al. [24] assessed the diagnostic performance of O-RADS MRI to characterise ultrasound indeterminate adnexal lesions, reporting high specificity and sensitivity. They reported higher malignancy rates for two of the five O-RADS MRI categories (scores 4 and 5) as compared with the predicted ones. Their finding agreed with a multi-centre European study [25], that reported high misinterpretation rate for these two O-RADS scores. Both studies revealed the need for quantitative biomarkers for lesion discrimination for ultrasound indeterminate ovarian lesions.

Table 2.

Studies assessing MR biomarkers in ovarian cancer, derived by O-RADS MRI functional techniques

Author	Year	O-RADS MRI Quantitative technique	Type of tumours	Aim of the study	Quantitative analysis/ assessed biomarkers	Study Outcome
Li	2012	DWI	Ovarian surface epithelial tumors	Differentiation between benign and malignant ovarian tumors preoperatively.	ADC	ADC significant different between malignant and benign.
Lindgeren	2017	DWI	Ovarian cancer	Utility of ADC as histological and prognostic marker.	ADC	i) Association between ADC and histological severity. ii) High prognostic value.
Wang	2017	DWI	Epithelial ovarian cancer	Utility of histogram analysis of mono-, bi-, and stretched-exponential models.	Histogram analysis of i) Monoexponential (ADC), ii) Biexponential (D, D*, f) iii) stretched (DDC, α) exponential models.	High diagnostic performance of diffusion coefficients of all models
Li	2017	DCE	Complex ovarian tumors	Differentiation of malignant, borderline, and benign complex ovarian tumors.	Semi-quantitative analysis: i) TIC ii) EA iii) MS iv) THR.	i) Useful for differentiation malignant from benign and borderline from benign. ii) It couldn’t differentiate between malignant and borderline.
Pi	2018	DWI	Ovarian tumours	Differentiation of benign, malignant and borderline tumors.	Meta-analysis of ADC	i) High sensitivity and specificity in differentiating malignant from benign. ii) High sensitivity and moderate specificity in differentiating borderline from malignant.
Lindgren	2019	DCE	Primary and metastatic ovarian cancer	Assessment of prognostic and predicting value of DCE parameters.	Semiquantitative analysis: i) Area under the curve, ii) WashIn iii) WashOut iv) Peak v) Time to peak. Quantitative analysis: i) Ktrans ii) Kep iii) Ve iv) Vp	Ktran beneficial in ovarian cancer prognosis.
De Piano	2019	DWI and DCE	Ovarian cancer	Assessment of prognostic value of DWI and DCE parameters.	DWI: i) ADC of ovarian masses (O-ADC), ii) ADC of ovarian mass normalized to the muscle (M-ADC) DCE: i) BAT area ii) CER area iii) IAUGC) iv) Kep v) Ktrans vi) MSL vii) Ve viii) INT	i) A slightly significant association between normalized ovarian ADC and the presence of residual tumor at surgery. ii) The other perfusion and diffusion parameters were not significant.
Li	2020	DCE	Epithelial ovarian cancer	Assessment of the prognostic value of whole solid tumor volume histogram parameters measured from quantitative DCE-MRI.	Quantitative analysis: i) Ktrans ii) Kep iii) Ve	Whole solid tumor volume quantitative DCE-MRI histogram parameters could i) predict the recurrence of HGSOC ii) not predict the recurrence of EOC.
Türkoglu	2020	DWI and DCE	Benign and malignant ovarian masses	Discrimination between benign and malignant ovarian masses.	DWI: i) b800 SI ii) ADC DCE: i) Ktrans ii) Kep iii) Ve	i) ADC, Ktrans, Kep could differentiate between benign and malignant ovarian masses. ii) Ve could not differentiate between benign and malignant ovarian masses.
He	2020	DWI and DCE	Borderline and malignant epithelial ovarian tumors	Differentiation between borderline and malignant tumours utilising whole solid-tumor volume histogram analysis of DWI, DKI, IVIM, and DCE-MRI.	DWI i) ADC ii) K iii) Dk iv) D v) D* vi) F DCE i) Ktrans ii) Kep iii) Ve	Diffusion-related parameters were more valuable for differentiation between borderline and malignant epithelial ovarian tumors, with the most reliable the 10th percentile of Dk.
Wei	2021	DCE	Malignant ovarian tumours	Evaluation of diagnostic efficacy of semi- and quantitative DCE-MRI for malignant ovarian tumours.	Meta-analysis of semi-quantitative and quantitative	Higher diagnostic performance of quantitative DCE as compared with the semi-quantitative analysis.
Wegnert	2022	DCE	Benign and malignant adnexal tumours	Comparison of the diagnostic accuracy of visual assessment with that of TIC assessment of DCE-MRI scans to categorize adnexal lesions as benign or malignant.	Image analysis with O-RADS score and TIC	TIC more accurate than visual assessment for achieving optimal diagnostic accuracy.
Derlakta	2022	DWI and DCE	Epithelial serous ovarian cancer	Description of DWI and DCE parameters in primary course, in disease course prognosis and treatment response.	DWI i) ADC DCE i)TTP ii) Perf. Max. En	i) Prognostic value of ADC of serous ovarian cancer. ii) ADC correlation with proliferation markers and angiogenesis. iii) TTP and Perf Max En correlation with PFS.
Hottat	2022	DWI	Malignant and benign adnexal masses	Evaluation of the ADC and ADC histogram value after segmentation of the whole-lesion in the O-RADS MRI score.	DWI i) ADC of ROI in a lesion ii) ADC histogram of whole lesion	ROI ADC and whole-lesion ADC histogram means of malignant lesions were significantly lower than those of benign lesions.
Assouline	2022	DWI	Adnexal masses with cystic components	Test and investigation of new specific MRI features related to cystic components to improve the ability of the O-RADS MRI score to stratify lesions according to their risk of malignancy.	ADC	Cystic component analysis may improve the diagnosis performance of the O-RADS MRI score in adnexal cystic masses.
Manganaro	2023	DWI	Adnexal masses	Introduction and evaluation of DWI and ADC in O-RADS MRI and assessment of any change in diagnostic performance.	ADC	Better diagnostic performance of O-RADS/ ADC MRI score than the classical O-RADS MRI system.
Bourourou	2024	DWI	Adnexal masses	Evaluation of ADC contribution analysis of the solid tissue of adnexal masses for tumour characterisation and risk stratification of the O-RADS MRI 4 category.	ADC	ADC analysis of solid tissue of adnexal masses could help distinguish invasive masses within the O-RADS MRI 4 category.

O-RADS: Ovarian-Adnexal Reporting and Data System, DWI: Diffusion weighted imaging, DCE: dynamic contrast enhancement, ADC: apparent diffusion coefficient, D: true diffusion coefficient, D*: pseudo-diffusion coefficient, f: perfusion fraction, DDC: distributed diffusion coefficient, α: intravoxel water diffusion heterogeneity, TIC: time-signal intensity curve, EA: enhancement amplitude, MS: maximal slope, THR: time of half rising, Ktrans: volume transfer constant from vascular space to extravascular and extracellular space, Kep: rate transfer constant of gadolinium from the extravascular to the vascular space, Ve: volume of extracellular and extravascular space, Vp: plasma volume, BAT: Bolus arrival time, CER: contrast–enhancement ratio area, IAUGC: initial area under the gadolinium concentration time curve, MSL: maximum slope of increase area, INT: intracellular space, EOC: epithelial ovarian carcinoma, HGSOC: high-grade serous ovarian carcinoma, b800 SI: signal intensity of a b-value of 800 s/mm², DKI: diffusion kurtosis imaging, IVIM: intravoxel incoherent motion, K: diffusion kurtosis, Dk: diffusion coefficient, TIC: Time-Intensity Curve, TTP: time to peak, Perf. Max. En.: perfusion maximum enhancement, PFS: Progression free survival

In the last decade, quantitative metrics of DWI and DCE-MRI have been thoroughly evaluated for their use as imaging biomarkers in ovarian cancer [24]. In DWI, both the image intensity and the derived ADC value have been used for lesion classification [26]. Different approaches to ADC analysis have been described. ADC threshold studies, such as those by Lindgeren et al. [1] and Derlatka et al. [2], demonstrated correlations between ADC and histologic or prognostic markers, including proliferation markers (Ki67) and angiogenesis factors (VEGF). Bourourou et al. [27] and others [28, 29] proposed ADC thresholds for O-RADS MRI 4 lesions to improve discrimination. However, a validated ADC cut-off value for routine clinical use in ovarian cancer has not been established. ADC histogram analyses, which capture intra-tumoral heterogeneity, have shown stronger correlations with histologic factors than mean ADC values, enhancing lesion classification [30]. Combined ADC O-RADS approaches, such as modified O-RADS/ADC MRI scores [31] have been suggested to improve standardisation and characterisation of indeterminate adnexal lesions [28].

DCE quantitative parameters have also been investigated as imaging biomarkers in lesion characterisation and disease prediction. Li et al. [32] investigated the usefulness of DCE-MRI for differentiating between malignant, borderline and benign ovarian tumours. They suggested that semi-quantitative analysis of DCE-MRI could distinguished malignant from benign tumours and borderline from benign tumours, but it could not differentiate malignant from borderline tumours. The same group [33] also evaluated the utility of whole solid tumour histogram analysis of quantitative DCE-MRI for the prediction of epithelial ovarian carcinoma (EOC) recurrence. They reported high reproducibility of the whole solid tumour volume quantitative DCE-MRI histogram analysis, but these parameters could only predict the recurrence of patients with high-grade serous ovarian carcinoma (HGSOC) and not EOC. Another group [34] reviewed the prognostic and prediction value of DCE parameters in ovarian cancer. They concluded that DCE parameters, especially volume transfer constant (K_trans) could be beneficial in ovarian cancer prognosis, but they emphasised the value of this parameters’ combination with anatomical and functional imaging as prognostic markers. Assessing the semi-quantitative DCE parameters, Wegnert et al. [35] highlighted the diagnostic value of time-intensity curve (TIC) analysis for indeterminate lesion characterisation, revealing higher specificity as compared to visual assessment, when O-RADS MRI is used for diagnosis. A meta-analysis study [12] evaluated the diagnostic performance of quantitative and semi-quantitative DCE for malignant ovarian tumours. It was reported higher diagnostic performance of quantitative DCE as compared with the semi-quantitative. The authors noted that clinical implementation of quantitative DCE is challenging due to high spatial and temporal resolution requirements across scanners, therefore recommended a semi-quantitative approach for clinical practice.

Challenges in O-RADS MRI biomarkers

Despite the increased interest and potential value of quantitative DWI and DCE for ovarian lesion differentiation, they have not been routinely implemented. A lot of studies have documented the challenges of utilising ADC or DCE metrics as imaging biomarkers for lesion classification in oncology. Major obstacles are the lack of standardisation in the acquisition protocols, the different post-processing methods, the different approaches in region of interest (ROI) selection. The reliable use of quantitative DWI and ADC is arguable in lesion characterisation, either due to the large heterogeneity of ovarian lesion resulting in ADC overlap between them [27] and the challenges related to DWI. A meta-analysis study [3] assessed the diagnostic performance of the ADC values in ovarian tumours discrimination. They reported high sensitivity and specificity (91%) for the differentiation between malignant and benign ovarian tumours, but high sensitivity (89%) and moderate specificity (79%) between borderline and malignant tumours. The study also emphasised the lack of standardisation in DWI acquisitions among the different studies. Similarly, an editorial [36] highlighted the ADC variability across different MRI scanner manufacturers and different sites as an obstacle to clinical implementation. According to the author, this is due to the heterogeneity of imaging parameters chosen by the sites as well as differences between acceleration techniques used and ADC calculation from the manufacturers. Another source of variability could be the different approaches for the selection of regions of interest (ROI) used in the analysis, some researchers use the whole tumour to assess ADC while others limit it to a suspected area of the malignancy. This lack of harmonisation among centres and scanner manufacturers has limited the clinical value of ADC cut-off values for lesion categorisation [1] as different groups published different figures. Other groups [37] recommended the use of normalised ADC values or related ADC changes as an imaging biomarker to eliminate the ADC variability. Another factor undermining the reliable use of DWI as an imaging biomarker is its proneness to susceptibility artifacts [8]. These artifacts can be associated either with the motion of different organs in pelvis or bowel peristalsis, or with magnetic field inhomogeneities in the anatomic area. For example, the presence of metallic implants in the pelvic area will influence the image quality of DWI. In such cases, protocol optimisation is necessary to eliminate the metallic artifacts or to reduce the scanning duration. Similar challenges have been observed in the implementation of quantitative DCE as an imaging biomarker. Protocol variations across the sites, such as different temporal resolutions, influence either the diagnostic performance of the semi-quantitative, or the quantitative analysis [24]. The semiquantitative analysis might be less complicated and more widely settled [12], but it relies on radiologist’s assessment and conception [31]. Although quantitative analysis is more objective, its complexity and time demands limit routine clinical use [12]. Lastly, resources, protocol, and scanner constraints may limit the feasibility of DCE-MRI and enhancement curve acquisition across sites [31].To address these challenges, alternative imaging biomarkers for ovarian lesion discrimination have been proposed over the past decade. Wang et al. [38] assessed the diagnostic value of various parameters derived from the mono-exponential, stretched and biexponential diffusion models in epithelial ovarian cancer. The authors performed histogram analysis across the whole volume of the tumour, aiming to determine which diffusion model parameters is the most reliable biomarker of cancer discrimination. They reported high diagnostic performance of apparent diffusion coefficient (ADC), pure diffusion coefficient (D), distributed diffusion coefficient (DDC) of the three models as compared to all the assessed parameters, noting the highest accuracy by the DDC estimated by the stretched exponential model. In another study [39], the capabilities of DWI and DCE-MRI for benign and malignant ovarian masses discrimination were investigated. For DWI, they assessed the ADC and the signal of the mean b-value of 800s/mm² (b800) DWI for lesion differentiation. For DCE, they assessed multiple quantitative parameters, like volume transfer constant (K_trans), rate constant (k_ep), and extravascular extracellular volume fraction (v_e). They reported good reproducibility of the DCE data and its utility to differentiate benign and malignant lesions. For DWI, they specified an ADC threshold (0.93 × 10^{− 3} mm² /s) for malignancy differentiation and the highest sensitivity of the b800 signal in lesion discrimination comparing with all the other quantitative parameters. He et al. [40] performed histogram analysis of DWI, diffusion kurtosis imaging (DKI), intravoxel incoherent motion (IVIM), and DCE-MRI for differentiating between borderline and malignant epithelial ovarian tumours (BEOTs vs. MEOTs). From the 90 assessed histogram metrics of the parameters derived by this techniques, 52 indicators from 8 parameters were statistically significant. These were the ADC, diffusion kurtosis (K), diffusion coefficient (Dk), D, pseudo-diffusion coefficient (D*), K_trans, k_ep, and v_e. They reported that valuable biomarkers for BEOTs and MEOTs differentiation were the diffusion-related parameters, with the most reliable the 10th percentile of Dk. These imaging biomarkers require advanced acquisition and strong post-processing. Advanced diffusion models demand long scan times and high-performance gradients, while quantitative DCE metrics similarly rely on powerful scanners, limiting clinical implementation. To date, no combination of metrics has yielded a reliable biomarker for lesion discrimination, indicating that information beyond cellularity and vascularity is needed for ovarian cancer diagnosis.

Finally, a common limitation across quantitative imaging studies is the aggregation of borderline and invasive malignant tumours into a single “malignant” category. Despite their distinct biological behaviour and clinical implications. This approach may inflate reported diagnostic performance and limits clinical applicability, particularly in scenarios where accurate stratification between borderline and malignant lesions is essential. As a result, the incremental value of quantitative O-RADS MRI biomarkers remains uncertain.

Magnetic resonance spectroscopy as an imaging biomarker

MRS is a non-invasive imaging technique that quantifies tissue metabolites by measuring proton resonance frequencies, which reflects the local molecular environment of protons and provides biochemical information complementary to conventional MRI imaging [41]. These metabolic profiles provide additional insights for classification of ovarian or adnexal lesions, whose classification is indeterminate in ultrasound and MRI. In MRS, instead of an image, a spectrum is acquired displaying the different frequencies of protons within different molecules within the voxel of interest. The combination of these frequencies can be attributed to a particular biochemical tissue composition, and as such this can differentiate normal and cancerous tissue. In other words, MRS is a non-invasive technique that reveals metabolic composition and cancer-related metabolic changes [42, 43].

MRS use in ovarian cancer began in the early 2000s, as summarized by year and conclusions in Fig. 1. The initial assessments were focused on deriving one metabolite as the appropriate imaging biomarker for ovarian lesion classification. Okada et al. [44] assessed the MRS peak arrangement in various female intrapelvic tumours, aiming to associate them with the different histological types of tumours. They reported lactate (Lac) and choline (Cho) as biochemical markers, in agreement with other cancers’ reported markers. Lac signal was higher in malignant tumours as compared to benign, as an indicator of anaerobic glycolysis. Cho, as an indicator of cell membrane metabolic activity, was noticed in solid parts of malignant and benign tumours. The small cohort (23 patients: 6 malignant, 17 benign) and the inclusion of non-ovarian tumours limited specificity. In 2002 [45] the MRS arrangement was explored but limited to a small number of adnexal lesions. As a malignancy indicator, they reported a lipid (Lip) peak at 1.3ppm due to methylene group in fatty acids. Stanwell et al. [46], assessed the use of 3.0T MRS in 14 ovarian lesions, introducing the Cho/Cr ratio as a marker of malignancy and also noting higher Lac levels in malignant lesions. Another group [47] evaluated 39 gynaecological lesions, investigating Cho peaks, but differences between malignant and benign tumours were not statistically significant.

Fig. 1.

Timeline of all the magnetic resonance spectroscopy (MRS) studies in ovarian cancer since 2001

The feasibility of a 3-dimensional (3D) MRS (chemical shift imaging (CSI)) at 1.5T was explored two years later [48] in ovarian tumours. The authors assessed the total Cho (tCho), as a malignancy indicator, in a small cohort of 23 ovarian masses. They observed a tCho peak in most malignant tumours, though occasionally in benign lesions as well. McLean et al. [49] used 3.0T MRS to quantify metabolites in 23 ovarian masses, noting Cho peaks in most primary and metastatic lesions; its absence in some was attributed to technical limitations causing very low or no signal.

In 2011 [50], N-acetylaspartate (NAA) at 2 ppm was investigated to distinguish mucinous from non-mucinous ovarian tumours, with higher peaks in mucinous lesions and a cut-off NAA value proposed for discrimination. Another study [51] at 3.0 T assessed multi-parametric MRI for treatment prediction, finding Cho associated with response prediction but not with monitoring treatment. The largest study to date [52] of 69 patients evaluated multiple metabolite ratios (Cho/Cr, NAA/Cr, Lac/Cr, Lip/Cr), with only Cho/Cr showing promise for lesion discrimination. Malek et al. [53] confirmed that Cho peaks were more frequent in malignant than benign adnexal lesions in a study of 46 lesions.

While Cho peaks or Cho/Cr ratios showed promise in many studies, their diagnostic performance was inconsistent. Multi-metabolite approaches highlight the biological relevance of metabolite patterns and their potential specificity for malignancy, although overlap with benign or inflammatory conditions may exist. Mansour et al. [13] analysed Lac, Lip, NAA, Cho, Cr and acetone (Ace), in 230 patients, proposing a Cho/Cr cut-off of 1.7. They further suggested that the presence of both Lac and Cho peaks could serve as a better marker for lesion differentiation. Another study [54] of 89 found several metabolite ratios useful, with NAA/Cho showing the highest performance for epithelial ovarian masses (Fig. 2). Takeuchi et al. [55] evaluated NAA and Lip peaks at 3.0 T for differentiating malignant from benign mucinous tumours. They found the Lip peak, an indicator of cell damage, outperformed NAA and proposed it as a malignancy marker. Finally, another study [56] showed that while Cho peaks had moderate diagnostic value, they added value in differentiating O-RADS MRI 4 lesions, identifying cases missed by DWI.

Fig. 2.

(A) MR spectrum of the solid component of the mass, showing a high Cho peak, superimposed with voxel location (Ma et al. 2019). (B) MR spectrum of the cystic component of the mass, showing a high Cho peak, superimposed with voxel location. Adapted from Ma et al. 2019 [54]

Current status in MRS

The initial results of the different metabolites, as imaging biomarkers, are encouraging, but their implementation in clinical studies is still reserved. A significant reason is the multiple assessed metabolites and their arguable results among the different groups. Figure 3 summarises all the evaluated metabolites in ovarian cancer for lesion discrimination. Cho [44, 46, 47, 49, 51, 54], Lac [44, 47, 54], Lip [45, 47, 50, 52, 56], and NAA [52, 55, 56], peaks were suggested as malignancy indicators, but without a well-defined threshold for discrimination. In addition, except for a few studies [52, 54] that included more than 50 of patients, all the others assessed less than 50 lesions (Table 1). This small number of analysed data undermines their significance in metabolite biomarker roadmap.

Fig. 3.

Assessed metabolites per study

Another reason for the limited use of MRS is the lack of a clinical methodological consensus in acquisition and in data analysis. Each group utilises its own approach in scanning (Table 3), in spectra and in post-processing analysis. A variability is observed in the volume of interest (VOI) localisation, in the sequence parameters, the scanning set-up, the model for analysing the data and the methodology for assessing the metabolites’ concentration. Apart from these, not all the studies provided details about the acquisition and analysis.

Table 3.

Magnetic resonance spectroscopy (MRS) acquisition characteristics and parameters of the research studies in ovarian cancer

Author/ year	Magnetic Field Strength (T)	Localisation strategy	MR Sequence	TE (ms)	Voxel size	ROI location
Okada 2001	1.5	SVS	PRESS	135	18–27 cm3	Centre of the cystic or solid part
Cho 2002	1.5	SVS	STEAM	30	8 cm3 (2 × 2 × 2 cm3)	Adnexal lesion
Stanwell 2008	3.0	SVS	PRESS	135	N/A	Cystic and solid components
Booth 2009	3.0		PRESS	72	5.3 to 81.3 mm3	Solid part of the tumour
Mc Lean 2009	3.0	SVS	PRESS	144	N/A	Primary mass
Esseridou 2011	1.5	CSI	PRESS	135	80 × 80 × 80 mm3	Solid part of the tumour
Takeuchi 2011	3.0	SVS	PRESS	144	8 cm3 (2 × 2 × 2 cm3)	Ovarian tumour
Sala 2012	3.0	SVS	PRESS	144	N/A	Primary
Ma 2014	1.5	SVS	PRESS	135	8 cm3 (2 × 2 × 2 cm3)	Mass
Malek 2015	3.0	SVS	PRESS	135	2 to 12 cm3	Enhancement component
Mansour 2019	1.5	SVS	PRESS	120	N/A	Solid part of the tumour
Ma 2019	1.5	CSI	PRESS	135	1 cm3 (1 × 1 × 1 cm3) to 8 cm3 (2 × 2 × 2 cm3)	Whole mass
Takeuchi 2021	3.0	SVS	PRESS	144	8 cm3 (2 × 2 × 2 cm3)	Septa or solid components
Li 2021	3.0	SVS	PRESS	35	1 × 2 × 2 cm3	Solid part of the lesion

SVS: single voxel spectroscopy, CSI: chemical shift imaging, PRESS: Point resolved spectroscopy, ROI region of interest

No clear preference for magnetic field strength has been reported, with both 1.5 T and 3.0 T used; 3.0 T offers higher SNR despite greater field inhomogeneities [42, 47, 49]. Most studies (13/14) used Point RESolved Spectroscopy (PRESS) for acquisition with Single Voxel Spectroscopy (SVS). Only two groups acquired multi voxel spectroscopy (MVS) or CSI, one 3d [48] and the other 2d [54]. One study [13] used both SVS and MVS to include complex masses. Due to intrinsic low signal in MRS, SVS provides more reliable spectra due to low intrinsic MRS signal, whereas CSI allows larger-area metabolite assessment. Mainly, echo time (TE), scanning duration and voxel size were cited. Most studies used long TE (135ms and 144ms) and only three used lower ones (30ms [45], 35ms [56], 72ms [46]. Half of the studies used a TE of 135ms, aiming to eliminate signals from Lip and water [47] and to allow the Lac detection [44]. The rest of the studies [49–51, 55] used a 144ms TE. Scanning duration was typically around 5 min [13, 52, 54, 55].

The voxel size was fixed in 5 studies [45, 50, 52, 55, 56] while it was adjusted depending on the size of the lesion in all the others [44, 46, 54]. Furthermore, the localisation of the VOI varied amongst the studies and was mainly placed by a radiologist. Some researchers placed it purely on the solid part of the tumours [46, 49, 52, 56] excluding many lesions [57], or placed it on the solid and cystic parts of the solid and cystic tumours respectively [44, 47, 48, 55]. In one study [53], the VOI was located on the enhancement part of the lesion and in another one [54] the MVS volume placed on the entire mass.

Most studies used scanner manufacturer software for analysis, with a few detailing all analysis steps [45, 47, 48, 54]. Only five studies [49, 51, 56] used LCmodel [58]. Post-processing approaches included qualitative (visual assessment [48], semi-quantitative (3-class metabolite classification [44, 47, 50, 55, 57], and quantitative analyses (mainly with LCModel [49, 51, 55, 56], with Takeuchi et al. [55] using both. Absolute metabolite quantification is lacking due to absent internal standards, so relative concentrations are used, causing variability. Stanwell et al. [47] reported a Cho/Cr ratio equal or greater than 3.09, Ma et al. [52] reported a cut of value of 7.46, but Mansour et al. [57] a cut of value of 1.7 Cho/Cr ratio for differentiating malignant from benign lesions. This heterogeneity in acquisition and analysis limits clinical utility of MRS.

Overall, the current evidence is limited by heterogeneity in study design, acquisition protocols, and post-processing methods. Most studies are single-centre, retrospective, and include small cohorts with inconsistent metabolite reporting, limiting direct comparison and precluding robust meta-analysis. Consequently, current findings should be considered exploratory, and further large-scale prospective studies with standardised protocols are required to validate the clinical utility of MRS in ovarian lesion characterisation.

Technical challenges in magnetic resonance spectroscopy

MRS data suffers from low SNR, causing some studies to exclude significant portions of data. Acquisition- related challenges include motion artifacts from respiration and bowel movement [42, 43, 45, 49], local susceptibility effects, magnetic field inhomogeneity, and voxel mislocalisation. Only one group [49] used respiratory triggering, yet bowel motion still affected spectra. Field inhomogeneity, exacerbated by surrounding air and fluid, broadens metabolite peaks, making proper shimming essential. Post-processing challenges include improving SNR, for which filters like Gaussian have been used [47] although standardised denoising methods are lacking. To reduce contamination of the surrounding tissues, external volume suppression is valuable. External volume suppression, implemented in some studies [54], combined with accurate voxel placement, could reduce variability. A consensus is needed for both acquisition and post-processing, and collaboration with manufacturers is required to translate research tools into CE-marked/FDA-approved clinical software.

Future of MRS in ovarian cancer

Ovarian cancer is often diagnosed at advanced stages, characterised by an increased lesion heterogeneity [55]. There remains a clinical need for improved non-invasive differentiation of benign and malignant O-RADS MRI 4 lesions, which remain indeterminate on ultrasound and MRI. Despite extensive research of quantitative O-RADS MRI metrics, a reliable imaging biomarker has not yet been established.

MRS may provide insight into tissue biochemistry, providing metabolic information beyond the available from standard anatomical or functional imaging. However, its clinical and research adoption remains limited, due to low SNR, motion artefacts, magnetic field inhomogeneity, local susceptibility effects, and inaccuracies in voxel placement. Importantly, there is currently no standardized acquisition protocol or consensus guidance for shimming, voxel localization, denoising, or post-processing, which undermines reproducibility and cross-study comparison.

Despite these challenges, MRS may represent a promising complementary biomarker, that could contribute to improving the diagnostic performance of existing imaging biomarkers, particularly for indeterminate O-RADS MRI 4 lesions. To establish its diagnostic value, future studies should aim to harmonise acquisition protocols, standardise post-processing workflows and define robust metabolic markers or metabolite combinations. Community- endorsed consensus guidelines, as in brain MRS [57], will be important to ensure reproducibility, enable cross-study comparison, and support the evaluation of the clinical utility of MRS in ovarian lesion classification.

Author contributions

MVP: literature search, drafted manuscript, prepared figures and tables, MM: critically reviewed and suggested revisions. JM: critically reviewed and suggested revisions. KO: reviewed the manuscript and provided radiological expertise and critical feedback. AS: proposed the review topic, reviewed the manuscript, and provided radiological expertise and critical feedback.

Funding

Not applicable.

Data availability

Not applicable.

Declarations

Ethical approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.