Semi-automated histogram analysis of normal bone marrow using ¹⁸F-FDG PET/CT: correlation with clinical indicators

Whole-body ¹⁸F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) is a glucose metabolism-based imaging technique that is currently used for staging, restaging, and therapeutic evaluation of malignant tumors. PET is a diagnostic imaging modality with excellent quantitative properties and the various quantitative values that are calculated based on the standardized uptake value (SUV), which is a semi-quantitative value of FDG accumulation, are useful for distinguishing benign and malignant tumors and predicting their prognosis [1‐3].

In addition to malignant tumors, bone marrow (BM) FDG uptake could be affected by hematologic, immune, and inflammatory disorders, which cause increased cellularity and hyperactivity in the BM [4‐7]. BM FDG uptake has also been reportedly associated with white blood cell (WBC) count, serum C-reactive protein, and granulocyte-colony stimulating factor or erythropoietin [8, 9]. Although BM biopsy remains the gold standard for BM pathology diagnosis in clinical practice, BM biopsies are highly invasive and contain many sampling errors. Particular attention should be paid to the irregular distribution of local areas of active bone marrow and/or hematological malignancies, and thus, the lesions may be missed on random biopsies.

FDG-PET/CT has enabled less invasive visualization of glucose metabolism in BM throughout the body, which was considerably difficult with conventional morphological imaging such as CT and magnetic resonance imaging [10]. Recently, FDG PET/CT has been increasingly applied to the diagnosis of bone marrow failure such as myeloproliferative neoplasm, aplastic anemia, and myelodysplastic syndrome as well as malignant lymphoma and multiple myeloma [11‐14]. Some studies have reported that the region-of-interest was manually placed on certain bones and based on the semi-automatic measurement method for quantitative evaluation of BM [14, 15]. The segmentation of whole BM by a computational approach of PET/CT images has also been reported [16, 17]. However, the PET quantification method for BM, which was proposed in the previous studies, was complicated and has not been clinically applied. In addition, the quantitative values representing the FDG uptake pattern of normal bone marrow as a baseline have been undefined.

Therefore, the purpose of this study was to establish a standard for bone marrow FDG uptake patterns by a reproducible quantitative method with fewer steps using a commercially available automated image analysis software.

A total of 98 healthy adults (57 men and 41 women) were enrolled in this study (Table 2). The study participants included 5 (5 women), 18 (12 men, 6 women), 12 (8 men, 4 women), 15 (11 men, 4 women), 25 (14 men, 11 women), 16 (8 men and 8 women), and 7 (3 men and 4 women) participants in their 20 s, 30 s, 40 s, 50 s, 60 s, 70 s, and 80 s, respectively. The average and standard deviation of age was 55.5 ± 14.2 and 56.3 ± 18.9 for men and women, respectively, with no significant difference. The average LBM, and hemoglobin level were significantly different between men and women (p < 0.0001, Table 2). These results indicated that sex was a covariate, and the following univariate and multivariate analyses were performed for each group of men and women.

Extraction of BM PET from one subject took about 3 min with the automated process alone, and was completed within 5 min even if manual correction was required. The intersection over union for the extracted pixel accuracy and the intraclass correlation coefficient for the SUL_mean was conducted for the groups of men and women. The extracted BM PET had a mean intersection over union of 0.99, and a SUL_mean intraclass correlation coefficient of 0.99, indicating an excellent reproducibility. In 70 of the 98 cases (71.4%), the BM PET was completely extracted automatically. In the other 28 cases (28.6%) that required manual correction (due to high CT values of thyroids: 7, metals for dental treatment: 9, pelvic arterial calcifications: 4, artifacts of pelvic bone: 4, inseparable pelvic and femur: 2, metal for lumbar spine surgery:1, and metal for femur surgery:1), the mean and standard deviation of intersection over union were 0.9875 and 0.0245, respectively. Among the BM PET-derived parameters, the volume, %Higher volume, and Deauville 5-point score were significantly different between men and women (Table 3). The volume increased with increase in LBM in women (r = 0.39, p = 0.0108).

Although the physiological uptake of FDG in the spinal BM has been shown to vary depending on the location of the vertebrae [15], to reduce intra- and interobserver variabilities, the BM extraction of all vertebrae and pelvis was calculated as one organ in this study. Extracted BM images of PET, which almost corresponded to the medullary cavities of the spine and pelvis, were automatically extracted using CT and PET images, and various PET quantitative values were measured. The FDG uptake in BM computed in our study was lower than that in previous reports, even with the difference that our method was corrected for LBM [15, 23] for both sexes. This may be due to background subjective variations and the measurement of larger volumes of BM of areas with low FDG uptake (upper cervical spine, sacrum, and pelvic bone) in the present study, whereas previous reports measured certain bones.

SUL_max, SUL_mean, entropy, and % higher volume of men were higher than those of women, and the proportion of women with a high Deauville score was higher than that of men. These results suggest that although the glucose metabolism of the BM of women is lower than that of men, the bone marrow of women has a relatively high glucose metabolism than other organs. This may be due to the difference in body size between men and women and the presence or absence of menstruation. Since the distribution of normal bone marrow cells is heterogeneous compared to abnormal increases in cell density such as tumors, differences in entropy between men and women may be the same reason. Furthermore, our study showed that FDG uptake in the BM decreased with age in both men and women, it was consistent with the previous studies [21, 23]. Since all of the participants' WBC and hemoglobin levels were in the normal range, the effects of abnormally low and high levels of these hematological indicators on BM uptake could not be fully elucidated. Furthermore, the entropy was also evaluated as one of the histogram features computed from BM PET images. The entropy of BM PET was higher with younger age, indicating heterogeneity in both men and women. Patient-based studies on these findings are warranted.

van Son et al. reviewed previous reports on FDG uptake levels and patterns on FDG PET/CT of hemato-oncological conditions [24]. The review summarized that inflammation and colony-stimulating factors caused a diffuse FDG uptake in the bone marrow that was higher than the physiological FDG uptake in the liver. Diffuse FDG uptake in the BM has also been found in myeloproliferative neoplasms [11, 12], while hyperplastic hematopoietic islands in aplastic anemia showed island-like and localized FDG uptake [13, 25], and BM reconversion with island-like and localized FDG accumulation following the treatment of malignant tumors were observed. Although multiple myeloma had a relatively low diffuse uptake in BM, the CT often showed bone abnormalities such as osteolytic mass and fracture [24]. Bone marrow invasion of malignant lymphoma often presented with localized high uptakes and might also present with diffuse uptakes [24]. Although these findings greatly influence the diagnosis and follow-up of various diseases, quantitative methods have not been established. Therefore, the FDG-uptake pattern in the BM could be characterized by the histogram parameters demonstrated in this study, which represent the FDG uptake’s heterogeneity. The FDG-uptake pattern obtained from this analysis serves as a baseline for future analysis of BM in various hematopoietic disorders.

In this study, we used a commercially available imaging workstation that implemented a recently developed deep learning-based organ segmentation technique for medical imaging [19, 26]. It achieved a highly reproducible and easy extraction of BM PET images of the spine and pelvis from PET/CT images. Although some previously reported computational approaches for bone and BM segmentation from CT-based PET/CT images exist, no direct comparison between these methods and ours was made [16, 17, 27]. Similar to the SUV measurement of malignant solid tumors being widespread in the clinical application of PET, standardization of a simple and accurate quantification method of BM PET for hematopoietic disorders is also necessary.

The major problem of this paper was that there was an insufficient assessment of the mechanism that brings the results of each parameter for each sex. This study evaluated only the bone marrow of healthy adults. The mechanism of various FDG uptake patterns in the bone marrow and its PET parameters can be elucidated by a comparative analysis of abnormal bone marrow accumulation caused by various diseases in the future.

Recent studies have revealed the feasibility of thymidine analog ¹⁸F-fluorothymidine (FLT) PET for the diagnosis and monitoring of BM disorders [28, 29]. FLT is a surrogate marker for DNA synthesis and is an excellent PET tracer that can clearly visualize the proliferative activity of the BM; however, it is still not available for other purposes excluding research. Application of the BM quantification method presented in this study to FLT-PET in the future could lead to more useful PET examinations for various hematopoietic disorders.

Our study had some limitations. First, there may have been a selection bias because this was a cross-sectional study using real-world PET imaging data for cancer screening. In addition, the sample size of the survey was small, with less than five participants in their twenties. In addition, some blood data containing C-reactive protein, which is known to correlate with BM-FDG uptake, were not analyzed. Therefore, more participants need to be tested. The blood test did not include CRP or erythrocyte sedimentation rate, which are indicators of inflammation, because it was associated with cancer screening of asymptomatic participants. However, these inflammatory indicators are associated with various diseases and should be included as routine tests. Second, the extracted BM PET contained some bone tissue, and it was not possible to prove that it completely matched the medullary cavity for each individual. Although 2 pixels were removed from the margin on bone CT, the removal of the cortical bone was not perfect because the thickness of the cortical bone could vary by site, age, and sex. However, FDG accumulation in normal bone (rather than bone marrow) is extremely low that it seems that glucose metabolism imaging of bone marrow could be represented almost accurately. In addition, if the area of the spinal canal surrounded by bone was small, it could not be excluded, as shown in Fig. 2a. However, if it was large, it could be excluded, as shown in Fig. 2b. In this study, we did not add manual corrections to establish a simpler method of bone marrow extraction. In the future, it will be improved by applying an algorithm that can accurately extract bone by CT value. Third, the extraction method was not completely automatic and some artifacts were manually removed. This will cause poor reproducibility. CTs reconstructed using the metal artifact reduction algorithm will be useful for fully automated extraction in future studies. Fourth, displacement of PET and CT images due to body movements during PET/CT scans is a major concern. Correction of the PET/CT fusion images using artificial intelligence, which has progressed remarkably, could be possible in the near future.

We presented a simplified FDG-PET/CT BM quantification method and revealed uptake patterns in healthy adults using a histogram analysis. The SUL_mean and entropy were correlated with age in both men and women, and SUL_max was also correlated with age in women. These results could provide baseline findings for various BM abnormalities.