Introduction
Breast cancer is the most commonly occurring malignancy in women worldwide, representing 11.6% of newly diagnosed cancer cases in 2018 [
1]. Disease prognosis changes dramatically if breast cancer is diagnosed at an early vs later stage, with the 5-year survival rate decreasing from 98 to 100% for the former to 66–98% for the latter [
2]. Despite the many advantages offered by new surgical approaches and targeted drug development, early diagnosis remains one of the most effective means to conquer breast cancer.
Imaging modalities that are currently used to diagnose breast cancer include mammography, ultrasound, and magnetic resonance imaging (MRI) [
3]. MRI, which is based on the depiction of neoangiogenesis as a tumor-specific feature, is the most sensitive imaging modality for breast cancer detection. However, a challenge in the broader use of breast MRI is its false-positive findings which lead to unnecessary invasive biopsies in benign tumors, along with unnecessary financial costs and patient anxiety [
4]. Factors that affect MRI’s specificity include the image acquisition technique and the level of reader experience [
4].
Carcinogenesis is a complex, multistep process during which cancers develop distinct pathological biological properties, i.e., cancer hallmarks, including sustained proliferation; evasion of growth suppressors and apoptosis; and promotion of angiogenesis, invasion, and metastasis [
5]. Advanced imaging techniques that provide morphologic, functional, and metabolic information have been introduced, allowing the non-invasive depiction of these pathophysiological processes at the cellular level. These novel imaging data can be used for tumor diagnosis and characterization, assessment of treatment response, and prediction of patient outcome [
6].
Simultaneous multiparametric
18F-fluoro-2-deoxy-
d-glucose (
18F-FDG) positron emission tomography/magnetic resonance imaging (PET/MRI) is a novel imaging technique that combines multiparametric morphologic and functional information from MRI with metabolic information provided by PET, offering unique insights into tumor biology to achieve the ultimate goal of precision medicine in oncology [
7,
8]. Recent studies support the use of
18F-FDG PET/MRI in breast cancer patients for different diagnostic purposes [
9,
10]. Initial studies using the combination of separately acquired MRI and PET data indicate an improvement in the discrimination of benign and malignant breast lesions [
11]. However, at present, the role of simultaneous multiparametric
18F-FDG PET/MRI for breast cancer diagnosis has not been fully assessed.
Recently, a new paradigm in healthcare has emerged, driven by advances in medical imaging technology and image analysis as well as the advent of artificial intelligence (AI) and its applications in medical imaging. Radiomics is the extraction of large numbers of quantitative features from standard-of-care medical images using computer algorithms; radiomics features can be correlated with various variables, e.g., patient characteristics and outcomes, and pooled in large-scale analyses to create decision support models [
12‐
14]. Radiomics has the potential to represent “the bridge between medical imaging and personalized medicine” [
15].
We hypothesized that an AI-based radiomics model combining quantitative simultaneously acquired 18F-FDG PET/MRI data will enable an accurate discrimination of benign and malignant breast tumors. Therefore, the aim of our study was to develop and validate a diagnostic AI model using quantitative perfusion, diffusion, and metabolic data as well as radiomics features extracted from simultaneous multiparametric 18F-FDG PET/MRI to non-invasively discriminate between benign and malignant breast lesions.
Discussion
At present, no studies have been published on simultaneous AI-enhanced 18F-FDG PET/MRI for breast cancer diagnosis. The aim of this study was to investigate whether an AI-based radiomics model combining quantitative simultaneously acquired 18F-FDG PET/MRI data enables accurate discrimination between benign and malignant breast tumors. A model including both quantitative parameters and radiomics features was shown to accurately discriminate between benign and malignant breast lesions. Our results indicate that AI-enhanced functional and metabolic breast imaging had excellent performance and outperformed expert readers, thus having the potential to assist human readers in correctly classifying suspicious breast lesions and obviate unnecessary invasive breast procedures.
While DCE-MRI is undisputedly the most sensitive test for breast cancer detection, with a pooled sensitivity of 99% [
34], there is still room for improvement in its diagnostic accuracy due to factors including overlap in imaging features between benign and malignant breast tumors, interpretation-influencing physiological factors such as background parenchymal enhancement, and last but not least human detection or interpretation error [
35].
To compensate for these limitations, additional functional and metabolic imaging techniques such as DWI, perfusion imaging, and PET have been developed that provide insights into tumor biology and thus improve diagnostic accuracy. Several studies have shown the incremental diagnostic value of these individual parameters [
36,
37]; particularly, their combined application as multiparametric MRI or PET/MRI has been shown to improve diagnostic accuracy for breast cancer detection and characterization [
11,
38].
Our findings also indicate that different functional and metabolic imaging techniques enable the non-invasive simultaneous depiction of oncogenic processes such as induction of neoangiogenesis, metabolic reprogramming, and sustained proliferation. In our study, the clinical interpretation of
18F-FDG PET/MRI showed good diagnostic accuracy with an AUC of 0.868 for breast cancer diagnosis, in line with previous studies [
11,
38,
39].
To fully leverage the wealth of information provided by simultaneous multiparametric 18F-FDG PET/MRI, we aimed to develop and validate a diagnostic AI model using quantitative perfusion, diffusion, and metabolic data as well as radiomics features to non-invasively differentiate benign from malignant breast lesions.
The AI model with the best diagnostic accuracy was based both on radiomics features extracted from ADC and PET images as well as the quantitative parameters DCE (MTT) and DWI (ADCmean) of breast lesions, achieving an accuracy, sensitivity, and specificity of 94.8%, 95.3, and 94.3%, respectively. This indicates that in order to enable the most accurate breast cancer detection information on tumor cellularity, metabolism and permeability are desirable.
It is worth noting that the model based on quantitative parameters only (i.e., ADC, MTT, and SUVmax) also showed a good performance (accuracy of 93.2%).
Although the multiparametric
18F-FDG PET/MRI AI-based radiomics model performed best, its performance was not statistically different from the clinical interpretation by expert readers. It has to be noted, however, that while clinical interpretation achieved similar sensitivities (95.3% vs 100%), the multiparametric
18F-FDG PET/MRI AI-based radiomics model achieved a higher specificity (94.3% vs 73.7%), highlighting the potential of such a model to reduce false-positive findings and obviate unnecessary breast biopsies in benign breast tumors [
36].
Several studies have been published on the use of AI applied to MRI for breast cancer diagnosis, mainly aiming at increasing its relatively low specificity compared to its high sensitivity, with accuracy values ranging from 72.8 to 92.0% [
40‐
44]. Similar to our work, Zhang et al. also explored the possibility to improve the accuracy of the ML classifier combining radiomics features extracted from both morphological and functional DCE and diffusion kurtosis (DK) images of 207 histologically proven breast lesions. They found that the model based on radiomics features from T2-weighted, DKI, and quantitative DCE pharmacokinetic parameter maps had the best discriminatory ability for benign and malignant breast lesions (AUC of 0.921) [
40]. In another study, radiomics coupled with ML analysis applied to DCE-MRI, including both radiomics features and clinical data, also proved to be accurate in the characterization of subcentimeter breast lesions in 96 high-risk BRCA mutation carriers, with a diagnostic accuracy of 81.5%, which was significantly higher than qualitative morphological assessment with BI-RADS classification (AUC of 53.4%) [
44]. The usefulness of a multiparametric MRI approach was explored in a recent study by Tsarouchi et al. [
45]. DCE and DW images of 85 breast lesions were analyzed for the extraction of first-order and texture features for the assessment of image heterogeneity and breast cancer diagnosis. Random forest resulted in the best performing algorithm (accuracy of 91.67%), combining both DCE-MRI and DWI parameters in a multiparametric assessment [
45].
Regarding PET imaging, the role of this functional technique has been explored in breast cancer mainly for prognostic/therapeutic purposes, particularly in the early prediction of the response to neoadjuvant chemotherapy [
46‐
48]. In a recently published study, the usefulness of radiomics and ML applied to PET/CT to differentiate breast carcinoma from lymphoma was investigated in a small number of lesions (19 breast lymphoma and 25 breast cancer lesions) [
49]. Different predictive models were built using combinations of clinical data, quantitative parameters (SUV), radiomics features (first- and second-order parameters extracted from both PET and CT images), and CT images. Models based on clinical data, SUV, and PET radiomics features as well as on clinical data and CT radiomics features were those that were most accurate (AUC of 0.806 and 0.759 in the validation cohort, respectively) [
49]. In an experimental study by Vogl et al. conducted on 34 breast lesions, a computer-aided segmentation and diagnosis (CAD) system was developed for automated lesion segmentation and classification (benign vs malignant) using separately acquired MRI and
18F-FDG PET/CT images [
50]. The CAD system achieved a Dice similarity coefficient of 0.665 for lesion segmentation and AUC of 0.978 for breast cancer diagnosis. While PET and DWI features improved DCE-MRI segmentation performance, such an improvement was not observed for lesion characterization [
50].
Limitations of our study have to be acknowledged. Firstly, our study is limited by the small sample size and the unbalanced distribution of benign and malignant breast lesions, with relevant implications for specificity. To overcome the limitation of the relatively small sample size, especially in regard to benign lesions, we opted to perform internal fivefold cross-validation which has been proven to be robust in such cases [
51]. The unbalanced distribution of benign and malignant lesions is related to the fact that this study is conducted at a single tertiary care cancer center and to the inclusion criterion of only patients with BI-RADS 0, 4/5 lesions which provides the clinical indication for performing a breast
18F-FDG PET/MRI. We addressed this limitation by using a well-established adaptive synthetic sampling to balance the two classes. Another limitation is the lack of external validation of the proposed AI model, which may limit its generalizability. To date, there is only a limited number of centers worldwide that have clinical simultaneous PET/MRI scanners for breast imaging. Collaboration with a different institution to validate our models is in development. Furthermore, two dynamic sequences were acquired before and after an update to the clinical MRI protocol. However, acquisition parameters were similar before and after the update, and AI techniques are meant to be applied to images acquired with different acquisition protocols; indeed, this issue did not affect the level of accuracy of the ML classifier. Finally, several cases had to be excluded from the analysis as at least one among DCE-MRI, DWI, or PET images was not suitable for the extraction of quantitative parameters or for radiomics analysis, in order not to impair the reliability of our data. Despite this stringent exclusion criterion, and also considering the limited access to such an advanced imaging technique, an adequate number of breast lesions was included in the final study sample which allowed the achievement of a good performance in the AI discrimination task.
In conclusion, a simultaneous multiparametric 18F-FDG PET/MRI AI-based radiomics model was shown to accurately discriminate between benign and malignant breast lesions. Our initial data indicate that AI-enhanced functional and metabolic breast imaging has the potential to assist human readers in correctly classifying suspicious breast lesions and therefore obviate unnecessary invasive breast procedures. Larger multi-center studies are being planned to validate the multiparametric 18F-FDG PET/MRI AI-based radiomics model.
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.