Distinguishing Benign and Malignant Findings on [68 Ga]-FAPI PET/CT Based on Quantitative SUV Measurements

Aim/Purpose Fibroblast activation protein (FAP) is overexpressed by cancer-associated fibroblasts. However, activated fibroblasts have been shown to play a significant role also in certain benign conditions such as wound healing or chronic inflammation. Therefore, the current study aimed to identify whether FAPI uptake might differ between malignant lesions and benign conditions. Material and Methods We retrospectively analyzed 155 patients with various cancer types who received [68 Ga]-FAPI-04/02-PET/CT between July 2017 and March 2020. SUVmax, SUVmean, and lesion-to-background ratios (LBR) of FAPI uptake were measured in benign processes compared to malignant lesions (primary and/or 2 exemplary metastases). In addition, receiver operating characteristic (ROC) curve analysis was conducted to compare the predictive capabilities of semiquantitative PET/CT parameters. Furthermore, the sensitivity, specificity, optimal cutoff value, and 95% confidence interval (CI) were determined for each parameter. Results Benign lesions exhibited significantly lower FAPI uptake compared to malignant lesions (mean SUVmax benign vs. malignant: 4.2 vs. 10.6; p < 0.001). In ROC analysis, cutoff values of these lesions (benign vs. malignant) were established based on SUVmax, SUVmean, and LBR. The SUVmax cutoff value for all lesions was 5.5 and the corresponding sensitivity, specificity, accuracy, and AUC were 78.8%, 85.1%, 82.0%, and 0.89%, respectively. Conclusion Our aim was to systematically analyze the pattern of FAPI uptake in benign and malignant processes. This investigation demonstrates that FAPI uptake might be useful to differentiate malignant and benign findings due to different patho-physiological origins. Supplementary Information The online version contains supplementary material available at 10.1007/s11307-022-01759-5.


Introduction
Recently, novel and highly promising PET tracers targeting fibroblast activation protein (FAP) by small molecule FAP inhibitors (FAPIs) have been introduced [1].
[ 68 Ga]-FAPI offers an additional pan-tumor tracer with excellent diagnostic performance and theranostic potential to the current standard oncological tracer 2-deoxy-2-[18F] fluoro-d-glucose ([ 18 F]-FDG). One of the advantages of this new approach represents its independence from glucose metabolism and the overexpression of FAP in more than 90% of human epithelial cancer cells [1]. Concurrently, FAP ligands demonstrate a rather low uptake in non-malignant tissue, ensuring favorable tumor-to background ratios (TBR) in malignant lesions. Furthermore, sharp image contrasts can be achieved [2], particularly in liver, brain, and oral mucosa, which are characterized by moderate to high physiological FDG uptake. Therefore, novel diagnostic opportunities such as FAPI imaging are crucial to overcome challenges in these tumor entities that display unfavorable TBR with FDG PET/CT (i.e., hepatic or pancreatic cancers) [3].
FAP is a type II serine protease belonging to the dipeptidyl peptidase 4 family and has both post-proline dipeptidyl peptidase and endopeptidase activity [4,5]. FAP is one of the surface markers expressed by cancer-associated fibroblasts (CAFs), which are part of the stroma in various malignant tumors [6]. The tumor stroma consists of endothelial cells, the basement membrane, extracellular matrix, and immune cells. In epithelial malignancies such as breast, colon, and pancreatic cancer, CAFs account for about 90% of the tumor mass [7]. CAFs secrete a number of molecules, mostly growth factors and cytokines (a prominent one being the transforming growth factor β (TGFβ)), which are shown to promote epithelial to mesenchymal transition [8]. Thus, FAP overexpression is associated with tumor immunomodulation leading to increased local tumor invasion, metastasis, and overall poor prognosis [6].
However, FAP is also selectively expressed in certain benign conditions and in normal tissues such as during wound healing [9], tissue remodeling [10], fibrosis, at sites of inflammation like arthritis [11,12], in atherosclerotic plaques [13], and after myocardial infarction [14]. Moreover, in immunohistochemical staining, elevated FAP levels are also present in the cervix and uterine stroma (especially during the proliferative cycle) as well as in the pancreas [15].
Thus, the distinction of incidental or confounding equivocal findings in the clinical setting appear to be one of the most frequent pitfalls. Although some data has already been published with respect to characterization of FAPI findings in benign tissues [16][17][18], predictive criteria for further clarification of such findings have not been described in the literature, yet. The aim of this work is to define or develop semiquantitative parameters that allow accurate prediction between benign and malignant FAPI uptake.

Patients
In this study, we retrospectively analyzed 155 oncological patients aged 31 to 93 with various malignancies (Table 1). Patients were referred by their attending oncologist or radiation oncologist and the initial evaluation of the data was conducted in a clinical setting at the University Hospital Heidelberg. The further statistical analysis was performed at the University Hospital Dusseldorf. Electronic patient records were additionally selectively analyzed to gather further information on clinical diagnoses. Several patients of this cohort were already evaluated regarding their malignancies under a different scope [2,[19][20][21][22], as well as their benign uptake in hormone-responsive organs including endometrium, breast, and ovary [23].
All patients gave written informed consent before undergoing [ 68 Ga]-FAPI PET/CT on an individual patient basis following national regulations, the Declaration of Helsinki and Good Clinical Practice (GCP). The radiopharmaceutical was synthesized and labeled under the terms of the German Pharmaceutical Act §13(2b). The retrospective evaluation of data was approved by the ethics committee of Heidelberg University (S-358/2022).

Radiopharmaceuticals and [ 68 Ga]-FAPI PET/CT Imaging
PET scans were performed in 3D mode with an acquisition time of 3-5 min/bed position (Supplement Table 1). Two FAP ligands with similar chemical compositions were used: FAPI-02 (n = 24) and FAPI-04 (n = 131). Radiosynthesis and labeling of 68 Ga-labeled FAP ligands were performed as described in previous publications [8,24]. Median injected activity was 239 MBq (range 118-340). Imaging data was acquired 1 h (n = 155) after tracer application. Whole body images, including the patients head to mid thighs, were acquired.

Image Analysis
[ 68 Ga]-FAPI uptake in tumor lesions and incidental/equivocal findings was quantified by mean and maximum standardized uptake values (SUV mean and SUV max ). Lesion-to-background ratio (LBR) was calculated by dividing the SUV max of the lesions (both benign and malignant) by SUV mean of blood pool to quantify the imaging contrast. Circular regions of interest (ROI) were placed by one UHD investigator (MD; supervised by FLG) on axial slices around lesions with focally increased tracer uptake in organs and tissues not affected by primary tumor or metastasis. The ROIs were then automatically incorporated into a 3-dimensional volume of interest (VOI) with e.soft software (Siemens) using a 60%-isocontour. The [ 68 Ga]-FAPI PET/CT scans were previously analyzed in consensus by a boardcertified radiologist, a board-certified radiation oncologist, and two board-certified nuclear medicine physicians.
Moderate to high organ or tissue tracer uptake and biodistribution were quantified by mean and maximum standardized uptake values (SUV mean and SUV max ). For comparison to benign FAPI uptake, SUV max and SUV mean of the primary (if present), blood pool and two representative metastatic lesions (if present) were measured as well. Evaluation of moderate to high FAPI uptake in organs and tissues not affected by malignancy was conducted with a 1-cm diameter (for the small organs and tissues like thyroid, esophagus, arteries, myocardium, oral mucosa, ovary, prostate) or 2-cm diameter (liver, pancreas, aortic lumen content, lung, mamma, endometrium, joints, wounds, intestines) sphere placed inside the organ parenchyma or tissue.
Primary lesions and metastases were defined considering imaging-based diagnosis, clinical diagnosis, or histopathological confirmation. Similarly, for the diagnosis of benign FAPI uptake, PET/CT imaging, data from other imaging modalities, medical history, and results of histological biopsy were taken into consideration. For further evaluation, tracer uptake of incidental or equivocal findings, blood pool, and tumor lesions were quantified by SUV max and SUV mean .
Regarding FAPI uptake, the lesions were categorized as low (SUV max < 1.5), moderate (SUV max 1.5-4) and high (SUV max > 4); the lesions with low uptake were excluded from statistical analysis and further evaluation.

Statistics
We used descriptive analyses for demographics, tumor characteristics, and tracer uptake. The descriptive analysis of the tracer uptake by SUV max , SUV mean , and LBR values between tumor and benign lesions were displayed using box plots (Fig. 2). Comparison of FAPI uptake between malignant and benign lesions was performed using the Mann-Whitney U test. A p value of < 0.05 was considered statistically significant. The descriptive statistical analyses were performed using Excel for Mac Version 15.41 (Microsoft) and Sigma-Plot 11.0 (Systat Software Inc.).
Receiver operating characteristic (ROC) curve analysis and the area under the ROC curve (AUC) were used to compare the predictive capabilities of semiquantitative PET/CT parameters regarding the differential diagnosis of equivocal findings and the sensitivity, specificity, optimal cutoff value, and 95% confidence interval (CI) were also calculated for each parameter. This statistical analysis was performed using MedCalc® Statistical Software version 20.011 (MedCalc Software). Optimal cutoffs were defined by Youden's index as those resulting in high sensitivity corresponding to highest negative predictive value or the maximum specificity for a given minimum level of sensitivity. The comparison of different AUCs was conducted by the method described by DeLong et al. [25]

Study Population
The cohort consists of 155 patients (70 females and 85 males) with a median age of 67 years (range 31-93 years) who underwent [ 68 Ga]-FAPI PET/CT scans between July 2017 and March 2020. Fifty different types of malignancies were included (Table 1). A total of 292 malignant lesion were evaluated, of which 82 were primary tumor lesions and 210 were metastatic lesions. The following cancer types were included: head and neck cancer (n = 27), lung carcinoma (n = 11), hepatobiliary cancer (n = 9), pancreatic cancer (n = 32), gastrointestinal tract cancer (n = 33), gynecologic cancer (n = 10), CUP (cancer of unknown primary, n = 6), prostate cancer (n = 9), and miscellaneous malignancies (n = 10) such as desmoid cancer, insulinoma, leiomyosarcoma, pheochromocytoma, thymic carcinoma and pancreatoblastoma. Fig. 1. The mean SUV max and median SUV max of the primary tumors were 12.1 and 11.5 (range 2.9-27.8), respectively. The mean SUV max and median SUV max of all metastases were 10.0 and 8.4 (range 2.4-44.3), respectively. Regarding the overall 618 benign lesions in 155 patients, FAPI showed a significantly lower uptake compared to malignant lesions (mean SUV max benign vs. malignant, 4.2 vs. 10.6; p < 0.001). Among the benign lesions, the uterine tissue (in particular, myometrium) demonstrated the highest FAPI uptake (n = 40, mean SUVmax: 9.9; p = 0.274) and represents the most outliers in the comparative statistical evaluation between benign and malignant lesions.

Discussion
This retrospective study primarily aimed to provide an overview of the frequency and clinical significance of incidental benign findings in [ 68 Ga]-FAPI PET/CT. Additionally, we sought to determine a reliable, accurate parameter or assessment method to facilitate imaging-based differentiation of those lesions. To this end, we analyzed a large patient cohort with a broad spectrum of histopathological verified cancer entities accompanied by confounding equivocal findings.
To distinguish between FAP expression of CAFs and the reactivation of the so-called quiescent fibroblasts in benign remodeling processes, i.e., wound healing and fibrotic changes, or physiological changes in hormone sensitive organs such as breast or uterus [26,27] is still a clinical challenge.
Studies with smooth muscle cells and formation of atherosclerotic plaques revealed a strong correlation between an increase of FAP expression and atherosclerotic changes in arteries which seems to be a poor prognostic factor because of the formation of thin-cap human coronary plaques [26]. In our evaluation, we observed 59 cases with moderately elevated FAPI uptake not only at sites of visible atherosclerotic plaques, but also in the thoracic and abdominal arteries as well as coronary arteries that showed no detectable signs Fig. 4 F A 60-year-old man with renal cell carcinoma showed a moderate, diffuse uptake of the testicles. As further side findings, BPH was noted. G A 76-year-old patient with esophageal cancer showing a diffuse, bilateral, high uptake of FAPI in the lungs, predominantly in the inferior lobe, fibrosis of the lung was diagnosed in the CT scan. Furthermore, this patient also showed moderate, diffuse uptake in the anal canal. No discernible pathologies were noted in the rectum or anal canal. H A 68-year-old-patient with colon cancer and two incidental findings; an intense uptake in the left lobe of the thyroid was noted. TSH levels were normal, and scintigraphy was recommended. Furthermore, the patient showed a moderate uptake in the middle and lower third of the esophagus without discernible esophageal pathologies in the CT scan or medical history. of arteriosclerosis. Presumably, the macrophage infiltration due to a subclinical atherogenesis leads to an increased FAP expression by TNFα-secretion [28]. However, this hypothesis requires further research to determine whether a decent increase of FAPI uptake within the vessel walls is a suitable predictive/prognostic factor for the atherosclerotic plaque formation and development of clinical manifestations.
As for FAPI uptake of the testicles, it has been established that seminiferous tubules are surrounded by the lamina propria, which consists of the basement membrane and the outer layers of myoid cells, connective tissue, and fibroblasts, whereas the outmost layers consist of mainly fibroblasts [29]. The composition of the peritubular wall can undergo dramatic changes. A study suggested that tryptase produced by peritubular mast cells may be directly involved in the development fibrosis, in particular the fibrotic thickening of the walls of the seminiferous tubules, causing infertility [30]. It might be possible, considering the advanced age of most male patients included in our study, that moderate FAPI uptake in the testicles might be associated with the these changes.
Among the benign lesions shown on [ 68 Ga]-FAPI PET/ CT, we observed 50 cases with moderate to diffusely high FAPI uptake in the pancreas. Some of these patients have had a prior cancer surgery, e.g., partial resection or postoperative pancreatitis as surgical complication. In other cases, without a surgical history and no clinical information suggesting acute pancreatitis, this appears to be related to chronic pancreatitis or fibrotic changes within the parenchyma. Furthermore, 25 patients were identified demonstrating moderate FAPI uptake at the site of active wound healing after operation (i.e., laparotomy). Overall, the diagnosis of an early tumor recurrence at sites of wound healing might be challenging due to high background activity of the target organ during the early postoperative period.
Several studies report an upregulation of FAPI in patients with thyroiditis and fibrotic changes in the thyroid parenchyma whereas FAPI uptake in differentiated thyroid cancers remains low [19,21]. We had a confirmed case of Hashimoto's thyroiditis in a patient with a diffuse increased FAPI uptake. Yet, 75 patients without history of thyroid pathologies demonstrated a diffuse or focal, moderate FAPI uptake in the thyroid parenchyma. Diffuse uptake of the thyroid might be linked to fibrotic changes of the thyroid architecture (e.g., lymphocyte and fibroblast infiltration) due to undiagnosed autoimmune diseases, like Hashimoto thyroiditis [31], Graves' disease [32], or immune-related thyroiditis caused by immunosuppressive therapy [33]. A recently published study supports this assumption [16].
Similarly, in 33 cases or 42% of male patients of our cohort, we noticed diffuse moderate uptake of FAPI in the prostate. The prevalence of prostatitis is estimated to range from 2.2 to 9.7% [34]; hence, a fourfold increase of incidental prostatitis seems very unlikely. Moreover, FAP expression in the benign prostate is shown to be relatively low. However, several studies report a crucial role of inflammation for the development and progression of benign prostatic hyperplasia (BPH) which has a high prevalence of more than 50% after 6th decade [35]. Hesterberg et al. demonstrated that FAP expression is significantly increased in the tumor microenvironment adjacent to Gleason grade 4 prostate cancer compared to benign prostate. Therefore, even a moderate, focal uptake of FAPI in the prostate should initiate diagnostic work-up in order to rule out prostate cancer [36] and an even distributed moderate to high FAPI uptake in the prostate might be assigned to BPH.
In concordance with previously conducted studies, we also noted a high, diffuse FAPI uptake in patients with fibrotic lung (12 cases) and liver diseases (9 cases). The application of FAPI for the diagnosis of primary or secondary hepatic malignancy is expected to be a promising diagnostic tool because of very low background uptake in liver tissue and thus excellent target-to-background ratio. However, the primary hepatic malignancies usually emerge in patients with fibrotic changes or advanced cirrhosis, which still might act as a limiting factor [37]. Moreover, intrahepatic expression of FAP seems to correlate with the degree of fibrosis [38]. Our evaluation seems to support these findings, as Fig. 3 shows a patient In the receiver operating characteristic (ROC) curve comparison, most favorable prediction rates (AUC > 90%) is seen in SUV mean with liver cirrhosis demonstrating high FAPI uptake in the liver. In addition, the conditions with chronic inflammation or infection in the lung, such as bronchiectasis, rheumatological disorders with pulmonary manifestations, restrictive lung disease or infections might represent, among others, a limiting factor for pulmonary tumor diagnostic.
Most of our patient cohort showed a moderate uptake in the rectum without any known underlying pathology. Zheng et al. reported of this phenomenon and postulated that mild fibrous tissue hyperplasia associated with varicose veins and the inflammation as a cause of increased FAP expression [18]. Undoubtedly, persistent straining, pressure, and microinjuries in the epithelium of the anal canal might cause a certain kind of the so-called never healing wound which, as we know, leads to a stronger FAP expression. Moreover, the increased glucose metabolism that is often detected by high FDG uptake also supports the hypothesis of chronic inflammatory changes of the epithelium in the anal canal. However, to our knowledge, there is no significant literature data supporting the correlation between fibrous tissue hyperplasia in varicose veins and FAP expression. Similarly, in 57 out of 155 patients, we identified moderate uptake in the esophagus, predominantly in the lower third of the esophagus. Gastroesophageal reflux disease (GERD) is one of the most predominant gastrointestinal diseases with a prevalence in the US between 18.1-27.8% and 8.8-25.9% in Europe (an even higher incidence is to be assumed due to an availability of over-the-counter anti-acidic medication) [39,40]. Studies suggest that cell damage and inflammation caused by gastric fluid may trigger the wound healing process and subsequent immunomodulation, promoted by inflammatory cytokines and chemokine activation [41]. Inflammatory cytokines cause fibroblast to myofibroblast transition, leading to fibrosis [42].
Beyond the characterization of benign findings in this study, we elaborated on the equivocal findings and classified them into benign and malignant lesions according to their etiology, which was determined by analyzing and crosschecking the electronic medical records including available pathology reports or reports of other imaging modalities. To our knowledge, a reliable parameter, or a diagnostic criterion such as a cutoff value for FAPI uptake is still missing. Our aim was to identify a pattern of FAPI uptake in benign lesions that could reliably predict the accurate diagnosis of equivocal findings in the clinical setting and thereby largely prevent misdiagnosis and mismanagement. Although there was sporadically overlap in the tracer uptake between benign and malignant lesions in some patients, we detected statistically significant differences in SUV max , SUV mean , and LBR in accordance with previous studies.
We performed ROC analysis to evaluate the diagnostic performance of [ 68 Ga]-FAPI PET/CT by assessing the semiquantitative metabolic parameters (SUV max , SUV mean , and LBR) in differentiation of equivocal findings on [ 68 Ga]-FAPI PET/CT, which yielded an excellent trade-off between sensitivity and specificity. We found a sensitivity of 78.8% and specificity of 85,8% at a SUV max cutoff value of 5.5 under AUC of 0.89. Our evaluation showed better diagnostic accuracy at a SUV mean cutoff value of 3.3 with a sensitivity of 84.9% and a specificity of 85,3% under AUC of 0.916. The cumulative accuracy and negative predictive factor for all three parameters are 82,8% and 89,7% respectively. Furthermore, the comparison of the ROC curves yielded a favorable result for the SUV mean , which, however, can be regarded as equivalent to other parameters. However, since the SUV max and SUV mean are not always comparable between different PET centers without cross-calibration of scanner hardware and reconstruction algorithms, the LBR might be a more reproducible predictive factor for the clinical and research setting as well. Since the FAP ligands, FAPI-02 and FAPI-04 are considered interchangeable due to comparable biodistribution at 1 h p.i. and similar diagnostic performance, no relevant influence on our analysis is expected [1].
Thus, our findings suggest rather promising, accurate diagnostic criteria for [ 68 Ga]-FAPI PET/CT distinguishing equivocal lesions based on semiquantitative metabolic parameters. We suggest that our findings can facilitate diagnostics by establishing predictable patterns of FAPI accumulation, especially while taking patient history into account. We have solely used semiquantitative parameters for the assessment, but if qualitative criteria such as the pattern and location of lesions, which can be evaluated only by a nuclear medicine physician, are also used, the diagnostic accuracy will probably increase further.
This investigation contained several limitations including its retrospective study design, with the associated risks of selection bias, since the cohort only consistsed of patients with known malignancies, frequently at advanced stages. Furthermore, the study is lacking a gold standard test for the validation of benign lesions. On the other hand, pathological analysis of every equivocal finding on [ 68 Ga]-FAPI PET/CT would be unfeasible and unethical. Finally, our results were obtained from a single institution and further multicenter studies are warranted to validate the utility of these metabolic threshold or cutoff values for differentiation of equivocal findings on [ 68 Ga]-FAPI PET/CT.

Conclusions
Since the implementation of [ 68 Ga]-FAPI PET/CT imaging, benign processes with significant FAPI uptake challenge accurate and precise distinction for clinical and research purposes. Within this investigation, we established that benign pathophysiological changes overall show a significantly lower uptake than malignancies, implying that the fibroblast activation in both settings may have distinct underlying pathophysiological mechanisms. Furthermore, we aimed to predict semiquantitative metabolic cutoff values with SUV max , SUV mean , and LBR to propose a differentiation of benign and malignant findings in a representative patient cohort with various malignancies.