Quantitative MRI Assessment of Bone Marrow Disease in Myelofibrosis: A Prospective Study

Abstract

Purpose

To evaluate quantitative MRI parameters for assessing bone marrow composition and fibrosis in individuals with myelofibrosis (MF), as a noninvasive alternative to biopsy.

Materials and Methods

This prospective, single-site study (ClinicalTrials.gov identifier no. NCT01973881) included participants with MF and with non-MF myeloproliferative neoplasms (MPNs) and healthy controls who underwent MRI scans from November 2016 to January 2024. Different MRI sequences assessed fat content (proton density fat fraction), cellularity (apparent diffusion coefficient, ADC), and cellularity/macromolecular structure (magnetization transfer ratio, MTR) across lumbar vertebrae, ilium, and femoral heads. The authors used linear discriminant analysis to classify the extent of bone marrow fibrosis for each participant based on ADC values.

Results

This study included 66 participants (45 with MF and 15 with other MPNs [34 female] and six healthy controls (four male)]. The median age was 63 years among participants with MF and other MPNs and 62 years among healthy controls. Participants in the MF subgroup showed elevated ADCs and MTRs with lower bone marrow fat than healthy controls. Individual bone marrow MRI metrics generally correlated across anatomic sites (Pearson r = 0.57–0.89). ADC in the ilium showed the highest correlation with pathologic grade of bone marrow fibrosis (Kendall τ_B = 0.44, P = .01). ADC values near the linear discriminant analysis threshold in two to three anatomic sites correlated with increased risk of overt bone marrow fibrosis (odds ratio = 5.81, P = .01).

Conclusion

Quantitative bone marrow MRI parameters, particularly ADC, correlated with bone marrow fibrosis and disease severity in MF.

Keywords: MR Imaging, Hematologic

Supplemental material is available for this article.

© RSNA, 2025.

Summary

MRI helped quantify macromolecular and cellular changes in the bone marrow of individuals with myelofibrosis, with increased apparent diffusion coefficient values correlating with biopsy-proven fibrosis.

Key Points

• ■ In a prospective study of 66 participants, including 60 with myelofibrosis (MF) or other myeloproliferative neoplasms (MPNs) and six healthy controls, MRI enabled noninvasive assessment of bone marrow composition as an alternative to biopsy.

• ■ Bone marrow fat content, cellularity, and macromolecular structure differed between healthy controls and participants with MF or other MPNs.

• ■ Increased apparent diffusion coefficient values in the ilium correlated with bone marrow fibrosis grade (Kendall τ_B = 0.44, P = .05), and when elevated across multiple sites, was associated with increased risk of overt fibrosis (odds ratio = 5.81, P = .01).

Introduction

Myelofibrosis (MF), a rare, ultimately fatal myeloproliferative neoplasm (MPN), can manifest as a primary disease or secondary to other MPNs such as essential thrombocythemia or polycythemia vera (1). Approximately 90% of MF cases arise from mutations in Janus kinase 2 (JAK2), calreticulin (CALR), or the thrombopoietin receptor (MPL) in hematopoietic stem cells. These mutations cause constitutive activation of the JAK/STAT pathway with dysregulated production of blood cells, elevated cytokines, and progressive bone marrow fibrosis (2). Disruption of bone marrow causes hematopoietic stem and progenitor cells to migrate to other organs, often the spleen and liver, where extramedullary hematopoiesis produces hepatosplenomegaly. Individuals with MF may experience debilitating constitutional symptoms such as fever, weight loss, and night sweats (3). Median survival varies widely, from approximately 1.5 to 14.2 years, based on risk groups defined by the Dynamic International Prognostic Scoring System (DIPSS). Mortality is attributed to causes such as infection and uncontrolled bleeding. Transformation to fatal acute myeloid leukemia occurs in 10% of patients (4).

Clinical management and clinical trials studying MF commonly rely on indirect measures of disease, such as spleen size and volume, peripheral blood counts, and scoring systems for constitutional symptoms to assess patients, owing to challenges in directly analyzing bone marrow in patients with MF. Biopsy of the marrow and aspiration from an untargeted site in the iliac crest are currently the only methods to analyze bone marrow cellularity, blast percentage, and extent of fibrosis (5), the latter of which pathologists grade on a semiquantitative scale from 0 to 3 (MF grade) (6). Interreader reliability of MF grade among pathologists is high, although inconsistencies in reporting may misclassify MPNs in up to 33% of individuals in registries (7). Inconsistent biopsy quality also may reduce diagnostic and prognostic utility (1). In patients with advanced MF, bone marrow biopsy and aspiration may produce sparse or no tissue, resulting in limited diagnostic value (8). Even with successful biopsies, recovered tissue represents a very small fraction of the total bone marrow, potentially obscuring the heterogeneity of disease features at different anatomic locations.

MRI may help overcome the limitations of biopsy and provide an advanced analysis of bone marrow in MF. Pilot studies with four participants demonstrated that proton density fat fraction (PDFF), a quantitative MRI metric for fat content, helped detect changes in bone marrow fat of patients treated for MF (9,10). Additional MRI sequences, including the qualitative assessment of T2-weighted images, and quantitative MRI parameters, including apparent diffusion coefficient (ADC) and magnetization transfer ratio (MTR), have shown promise for evaluation of bone marrow in mouse models of MF (11–14). ADC is the quantified output of diffusion-weighted imaging, which assesses how easily water moves within a tissue (15). Hypercellular bone marrow in MPNs with associated MF-related fibrosis increased ADC relative to the very low ADC in healthy bone marrow, which predominantly contains fat (16,17). Additionally, MTR values increased with greater volumes of macromolecular structures, such as cells and fibrosis, in bone marrow (18–20). We previously established that these MRI metrics strongly correlated with characteristic histologic changes of hypercellularity, normal bone marrow fat replacement, and fibrosis in mouse models of MF (12,13). We hypothesized that a combination of PDFF, ADC, and MTR would distinguish bone marrow profiles of healthy individuals from those of participants with non-MF MPNs and MF and reflect the heterogeneity of disease severity within the MF group. The aim of this study was to evaluate PDFF, ADC, and MTR as noninvasive, quantitative MRI parameters for assessing bone marrow composition and fibrosis in individuals with MF.

Materials and Methods

This single-site, prospective study (ClinicalTrials.gov identifier no. NCT01973881) was approved by the institutional review board of the University of Michigan and complied with the Health Insurance Portability and Accountability Act. All participants provided written informed consent.

Participants and Clinical Scoring

The pilot study included two arms, one for the quantification of MRI parameters for the mobility of water (diffusion, magnetization transfer) and/or fat in single-time-point scans from participants at any disease stage or duration to assess the extent of MF, and the other to measure changes in these parameters over longitudinal MRI studies (up to 2 years) in participants starting a new therapy for MF. Although the design of the pilot study did not include formal sample size or power calculations, we attempted to enroll similar numbers of participants for each grade of MF fibrosis as determined with biopsy. This study enrolled male and female participants older than 18 years with polycythemia vera, essential thrombocythemia, primary MF, postessential thrombocythemia MF, or postpolycythemia vera MF. The disease duration, treatment duration, and type of treatment (if any) were not specified as criteria for enrollment in arm 1, resulting in a heterogeneous cross-sectional cohort. The study also included six healthy participants with no known blood or bone marrow disease who served as age-comparable controls, close to the median age of 65 years among individuals diagnosed with MF. While the treatment monitoring component of the study is ongoing, we report herein the results from arm 1 for single-time-point examinations from participants who underwent MRI scans from between November 2016 and January 2024. We obtained clinical data including blood counts, mutational status, and treatment status/response by reviewing participants’ medical records and laboratory tests results obtained closest to the MRI scan date. Participants with MF were assigned a DIPSS risk group according to the International Working Group for Myeloproliferative Neoplasms Research and Treatment (21). For succinctness of presentation, we combined participants with either primary or secondary MF, recognizing that the DIPSS score may be more likely to assign a person with secondary MF to a high-risk group (22). A board-certified hematopathologist with 15 years of experience (W.L.) independently reviewed all bone marrow biopsies and determined the grade of bone marrow fibrosis according to the 2016 World Health Organization classification of MF (5). As the study did not require bone marrow biopsy or specify timing relative to the MRI, we present herein data for only participants with MPN who underwent biopsy within 1 month of MRI. If participants underwent both MRI and biopsy on the same day, we performed the MRI examination first.

Bone Marrow MRI and Generation of Quantitative Maps

We used a 3-T MRI scanner (Ingenia, software version 5.1.70; Philips Healthcare) with a 28-channel torso receiver coil and axial MRI scan geometry. We list detailed parameters for diffusion-weighted imaging, PDFF, and magnetization transfer in Appendix S1. For all three sequences, we utilized parallel imaging acceleration (sensitivity encoding factor, 2) in the anterior/posterior direction. PDFF and diffusion-weighted images included the lumbar vertebrae, pelvis, and femoral heads, while MT images included only the pelvis centered on the iliac crest to keep scan time < 1 hour.

We calculated ADC maps with the scanner using a monoexponential decay function for the ADC signal (23). MTR maps were generated in MATLAB version 2019b using (MT_off − MT_on)/MT_off. We automatically generated PDFF, in- and out-phase, and water- and fat-only images with the scanner, which included corrections for magnetic field inhomogeneity, T2* decay, and multiple peaks in the lipid spectrum (seven peak model).

Volume of Interest Generation and Image Quantification

We manually identified volumes of interest (VOIs) using 3D Slicer version 4.10.23. Because we used various acquisition parameters, we created separate VOIs for PDFF and ADC iliac and femoral head bone marrow. We used anatomic information from all the sequences to guide placement of the bone marrow VOIs on images with reduced contrast, particularly PDFF, in participants with overt MF-related bone marrow fibrosis. For PDFF and ADC VOIs in lumbar vertebrae, we manually identified the center of L1–L5 and used MATLAB to automatically construct three-dimensional ellipsoid VOIs in the bone marrow. For MTR images, we manually segmented iliac bone marrow and then identified regions of muscle and subcutaneous fat to normalize the MTR signal. To ensure consistency among participants, we quantified a vertical region (caudocranial direction) of at least 4 cm on ADC and PDFF maps for iliac and femoral head bone marrow. Because we acquired MTR scans across a narrow axial section, we evaluated all sections at a total approximate height of 2.8 cm. We used MATLAB 2019b to apply and quantify VOIs with MRI maps.

MTR Image Normalization

MTR values can vary based on sequence, vendor, and institutional differences (24); therefore, we present herein normalized MTR values to help generalize our results. Because fat and muscle are close to the bounds of the MT spectrum and readily identifiable in the acquired images, we normalized MTR maps as follows: (MTR_Acquired − MTR_Fat)/(MTR_Muscle − MTR_Fat). Hence, the reported MTR values distributed from 0 to 1, where 0 = MTR signal from fat and 1 = MTR signal from muscle.

Bone Marrow MRI Metric CIs and Repeatability Coefficients

We defined 95% CIs for each MRI sequence in each anatomic region to define approximate “healthy” and “MF” ranges. We utilized CIs that did not rely on repeatability coefficients, which are reported elsewhere for the test-retest scans of bone marrow PDFF and ADC for select anatomic sites in healthy volunteers (25–28).

Statistical Analysis

Quantifying the number of abnormal MRI anatomic region metrics

We analyzed data with GraphPad Prism 10 or MATLAB 2019b using P < .05 for statistical significance. We quantified bone marrow as the mean value of the VOIs from each anatomic region per MRI sequence. To classify participants based on ADC values, we first performed linear discriminant analysis (LDA) in MATLAB on vertebral body, iliac, and femoral head ADC data stratified by bone marrow fibrosis grade. We identified an LDA threshold for ADC at each anatomic site and determined for each participant if the corresponding ADC score for each anatomic site exceeded the LDA threshold for that site. This allowed us to rank participants by 0–1 or 2–3 ADC values above the LDA threshold. We evaluated the association between fibrosis grade and number of ADC metrics above the LDA using Fisher exact test and calculated sensitivity and specificity assuming a true-positive result showed 2–3 ADC values above the LDA threshold for participants with overt fibrosis (MF, 2–3) and a true-negative result showed 0–1 ADC values above the LDA threshold for those with no-to-early fibrosis (MF, 0–1).

Correlations

We analyzed continuous variables, such as comparisons of MRI metrics among anatomic regions, with Pearson correlation coefficient reported as r with statistical significance set at P < .05. Kendall τ_B was utilized for correlations involving ordinal-ordinal or ordinal-continuous comparisons reported between blood counts and the number of abnormal MRI metrics for each of the seven MRI-anatomy metrics and DIPSS, MF grade, and cellularity.

Other comparisons

We used Tukey multiple comparisons test to compare MRI metrics among anatomic sites or MF grades and an unpaired t test for cellularity between the non-MF MPN and MF groups.

Results

Participants

We analyzed 66 participants who underwent MRI scans between November 2016 and January 2024 (Tables 1, 2), 60 of whom had a documented MPN: polycythemia vera (n = 9), essential thrombocythemia (n = 5), MF (n = 45), or MPN/myelodysplastic syndrome (n = 1). We also enrolled six age-matched healthy controls with no known history of bone marrow disease or cancer. We included participants at any stage of disease or treatment, with approximately balanced numbers among the four MF grades. All recruited individuals participated in the study. The median age was 63 years (IQR, 51–73) among the participants with MPN or MF (34 of 60 [57%] female, 26 of 60 [43%] male) and 62 years (IQR, 58–65) among the healthy controls (two of six [33%] female, four of six [67%] male). Owing to the availability of healthy volunteers, there were more male than female controls. Most of the participants (67%) had JAK2 mutations, the most common cause of these malignancies (29). We conducted analyses on single-time-point data from participants at any stage of disease, duration of treatment, and type of treatment. As these inclusion criteria generated a heterogeneous cross-sectional cohort, we did not investigate the effects of specific driver mutations on the study outcomes.

Table 1:

Characteristics of Healthy Controls

Metric	Value
No. of participants	6
Age (y)*	62 (53–74)
Sex
Male	4 (67)
Female	2 (33)
Spleen volume (mL)*	195.07 (100.4–270.72)

Note.—Data are presented as numbers of participants, with percentages in parentheses, unless otherwise indicated.

^*Data are presented as medians, with ranges in parentheses.

Table 2:

Characteristics of Participants with Myeloproliferative Neoplasms

Metric	Value
No. of participants	60
Age (y)*	63 (30–88)
Sex
Male	26 (43)
Female	34 (57)
Spleen volume (mL)*	764.36 (140.42–5532.33)
Diagnosis
MF/Pre-MF	45 (75)
PV	9 (15)
ET	5 (8)
MPN/MDS	1 (2)
MF fibrosis grade
Grade 0	14 (23)
Grade 1	16 (27)
Grade 2	12 (20)
Grade 3	18 (30)
DIPSS risk category
Low	8 (18)
Intermediate-1	19 (43)
Intermediate-2	13 (30)
High	5 (9)
Transfusion dependence	10 (17)
Mutations
JAK2V617F	40 (67)
CALR	14 (23)
MPL	3 (5)
Triple negative	2 (3)
JAK2V617F + MPL	1 (2)

Note.—Data are presented as numbers of participants, with percentages in parentheses, unless otherwise indicated. DIPSS = Dynamic International Prognostic Scoring System, ET = essential thrombocythemia, MDS = myelodysplastic syndrome, MF = myelofibrosis, MPN = myeloproliferative neoplasm, PV = polycythemia vera

^*Data are presented as medians, with ranges in parentheses.

Characterizing Healthy, Non-MF MPN, and MF Bone Marrow Using Quantitative MRI

To assess MRI for detecting bone marrow abnormalities in MF, we evaluated lumbar vertebral body (L1–L5), iliac, and femoral head bone marrow using three quantitative MRI metrics: ADC (water diffusion, restricted by cells and extracellular matrix), PDFF (fat content), and MTR (macromolecular structure, such as extracellular matrix). These MRI sequences use endogenous tissue contrast and, therefore, do not require the use of exogenous contrast agents. MTR sequences were performed only for iliac bone marrow to keep scan time to approximately 45–50 minutes. We assessed seven combinations of MRI and bone marrow anatomic regions: PDFF of the lumbar vertebrae, ilium, and femoral heads; ADC of the lumbar vertebrae, ilium, and femoral heads; and MTR of the ilium (Fig 1A). We assessed approximately 150 cm³ of total bone marrow in each participant (Fig 1B), compared with the <1 mL collected in a typical bone marrow biopsy and aspiration (30). Bone marrow of the healthy controls had high contrast compared with the surrounding tissue, making all three anatomic regions easily observable (Fig 2A). For each metric, the contrast of the bone marrow with surrounding tissue decreased progressively from healthy controls to participants with non-MF MPNs and then those with overtly fibrotic MF (Figs 2B, 2C). Visual distinctions among these groups showed progressive reductions in PDFF and increases in MTR and ADC from healthy controls to overt MF.

Figure 1:

MRI helps assess bone marrow in multiple anatomic locations. (A) The imaging study captured data from bone marrow in three anatomic locations: lumbar vertebral bodies, ilium, and femoral heads. We reconstructed apparent diffusion coefficient (ADC) and proton density fat fraction (PDFF) maps in all three anatomic locations and magnetization transfer ratio (MTR) maps in the ilium, resulting in seven total MRI-anatomy metrics to evaluate bone marrow. Bone marrow in each anatomic location (yellow arrowheads) is identified on representative PDFF maps from a healthy 62-year-old male individual (created with BioRender [2025]; https://BioRender.com/w35xmi7). (B) We quantify bone marrow volumes from each region. Each dot represents the volume of quantified bone marrow from individual participants in each anatomic region. Numbers in each bar denote the mean quantified volume of bone marrow per region. ADC = apparent diffusion coefficient, FH = femoral head, IL = ilium, MTR = magnetization transfer ratio, PDFF = proton density fat fraction, VB = lumbar vertebral bodies.

Figure 2:

MRI reveals differences in bone marrow in healthy participants and those with non-MF MPN and MF. Panels display representative coronal ADC and PDFF reformatted images and axial MTR images for (A) healthy (62-year-old male) participant and participants with (B) MPN–polycythemia vera (57-year-old male) and (C) MF (67-year-old male). Red arrows identify bone marrow sites of interest in the lumbar spine, ilium, and femoral heads. ADC = apparent diffusion coefficient, MF = myelofibrosis, MPN = myeloproliferative neoplasm, MTR = magnetization transfer ratio, PDFFF = proton density fat fraction.

Quantifying bone marrow in multiple anatomic sites revealed that the MRI signals for each sequence generally correlated across different regions (Fig 3; PDFF: r_VB:IL = 0.86, r_VB:FH = 0.72, and r_IL:FH = 0.89; ADC: r_VB:IL = 0.86, r_VB:FH = 0.57, and r_IL:FH = 0.66, where r is the Pearson correlation coefficient, FH is the femoral head signal, IL is the iliac signal, and VB is the lumbar vertebral bodies signal, and P < .01 for all correlations). We compared bone marrow signals between healthy controls, participants with non-MF MPNs, and those with MF at each anatomic site. PDFF progressively decreased from healthy controls to participants with non-MF MPNs and MF (Fig 4A), while ADC increased in participants with MF compared with healthy controls (Fig 4B). For all groups, PDFF increased from lumbar vertebral bodies to the ilium and then femoral heads, consistent with the increased presence of hematopoietic marrow in the spine relative to the appendicular skeleton (31). MTR in the ilium increased progressively from healthy controls to participants with non-MF MPNs, followed by those with MF (Fig 4C). The group means for each bone marrow region per MRI sequence (Table 3) showed that healthy controls and participants with MF differed (vertebral body PDFF: P = .001, iliac ADC: P = .02, iliac PDFF: P = .006, iliac MTR: P = .005, and femoral head PDFF: P = .02). While mean values differed between the groups, the data for individual participants may have overlapped between the groups, particularly for PDFF. There was no evidence of a bias between the volume of analyzed bone marrow and reported MRI metrics, other than a modest correlation for ADC in vertebral bodies (Tables S1, S2).

Figure 3:

ADC and PDFF correlate among anatomic regions. Panels illustrate relationships among lumbar vertebral bodies, ilium, and femoral head bone marrow for (A) PDFF and (B) ADC. X- and y-axes show mean values for percentage of fat measured with (A) PDFF and (B) ADC (10⁻⁶ mm²/sec) in vertebral bodies and ilium, respectively, while the pseudocolor scale for symbols for each participant reflects mean (A) PDFF (% fat) and (B) ADC (10⁻⁶ mm²/sec) values in femoral heads. Pearson correlation coefficients, r, for each MRI metric across anatomic regions are PDFF: r_VB:IL = 0.86, r_VB:FH = 0.72, r_IL:FH = 0.89; ADC: r_VB:IL = 0.86, r_VB:FH = 0.57, r_IL:FH = 0.66; P < .01 for all correlations. Analyses include all participants with non-MF MPN and MF, with ADC and PDFF data from all three anatomic regions. PDFF is given as percentage of fat. ADC is in units of × 10⁻⁶ mm²/sec. ADC = apparent diffusion coefficient, FH = femoral head, IL = ilium, MF = myelofibrosis, MPN = myeloproliferative neoplasm, PDFF = proton density fat fraction, VB = vertebral bodies.

Figure 4:

MRI helps detect differences in bone marrow among participants with non-MF MPN and MF and healthy participants. Graphs show quantification of (A) PDFF, (B) ADC, and (C) MTR from bone marrow of healthy participants and participants with non-MF MPN and MF. PDFF decreases, ADC increases, and MTR increases compared with healthy bone marrow. Individual data points represent mean bone marrow values for each participant, with the bars denoting the mean and error bars denoting the 95% CI of the sample mean. P values calculated using multiple comparisons test (Tukey method). * P < .05, ** P < .01, *** P < .001. ADC = apparent diffusion coefficient, MF = myelofibrosis, MPN = myeloproliferative neoplasm, MTR = magnetization transfer ratio, PDFF = proton density fat fraction.

Table 3:

MRI-derived Metrics for Each Participant Subgroup

Participant Groups and Anatomic Sites	ADC (×10−6 mm2/sec)	PDFF	MTR
Healthy
Lumbar vertebrae	437.2 ± 109.3	45.87 ± 9.69	…
Ilium	364.5 ± 65.6	59.72 ± 5.79	0.238 ± 0.157
Femur heads	369.0 ± 65.4	88.54 ± 3.06	…
Non-MF MPN
Lumbar vertebrae	411.8 ± 50.7	30.20 ± 10.11	…
Ilium	422.5 ± 76.3	42.49 ± 11.10	0.489 ± 0.122
Femur heads	411.8 ± 63.1	71.86 ± 14.38	…
MF
Lumbar vertebrae	550.8 ± 54.2	19.36 ± 4.99
Ilium	558.6 ± 52.4	31.50 ± 6.62	0.547 ± 0.063
Femur heads	452.8 ± 45.7	53.89 ± 9.05

Note.—Data are presented as means ± margins of error, defined as the half width of the 95% CI. ADC = apparent diffusion coefficient, MF = myelofibrosis, MPN = myeloproliferative neoplasms, MTR = magnetization transfer ratio, PDFF = proton density fat fraction.

Bone Marrow MRI Captured Relevant Disease Features of MF

As the number of abnormal MRI metric classifications increased, the magnitude of the distance from the healthy CI also increased for each metric (Fig 5). To assess the extent to which these MRI metrics captured disease manifestations of MF, we analyzed several parameters already used to evaluate MF: peripheral blood counts, spleen volume, prognostic score (DIPSS), and bone marrow biopsy data for cellularity and fibrosis grade. Peripheral blood counts (Table S3) showed only a modest correlation with the number of abnormal bone marrow MRI metrics (Kendall rank correlation: τ_{B_platelets:MRI} = −0.26, P < .05; τ_{B_ImmGranulocytes:MRI} = 0.14, P < .05; τ_{B_PeripheralBlasts:MRI} = 0.28, P < .05), while the total number of abnormal blood counts demonstrated a modest positive relationship with the total number of abnormal bone marrow MRI metrics (Fig S1, Kendall τ_B = 0.28). For participants with MF, spleen volume also correlated with several of the bone marrow MRI metrics, although wide CIs likely limit its clinical value (Table S4). To investigate the relationship between bone marrow MRI and prognosis, we compared bone marrow MR images for each participant with the corresponding DIPSS risk groups (Figure S2A–S2C). All three MRI metrics (ADC, PDFF, and MTR) correlated with DIPSS, although only iliac and femoral head ADCs demonstrated even modest correlation with the prognostic score (Kendall rank correlation: τ_{B_IliacADC:DIPSS} = 0.32, P = .0054; τ_{B_FemoralHeadADC:DIPSS} = 0.23, P = .0476). As DIPSS does not incorporate specific features of bone marrow disease in MF, such as fibrosis, we expected limited correlations (32); however, overall, these data imply that quantitative MRI revealed information about bone marrow status not fully captured with blood counts and spleen volume.

Figure 5:

Quantification of each MRI metric is illustrated for (A) vertebral body, (B) ilium, and (C) femoral head bone marrow in participants with an increasing number of abnormal MRI metrics. Each data point is the median value for each group of abnormal MRI metrics. Gray bands indicate the healthy 95% CI of the mean for each metric. PDFF is given as percentage of fat. ADC is in units of × 10⁻⁶ mm²/sec. MTR is a normalized ratio on a scale of 0–1. ADC = apparent diffusion coefficient, MTR = magnetized transfer ratio, PDFF = proton density fat fraction.

MRI Correlated with Relevant Features of MF Bone Marrow

We investigated the extent to which bone marrow MRI metrics across various anatomic sites correlated with cellularity and fibrosis as determined through bone marrow biopsy. We focused on ADC, PDFF, and MTR from the ilium, as it is the standard bone marrow biopsy site and was the only anatomic site from which we acquired MTR data. ADC, PDFF, and MTR all correlated with increasing MF grade, although ADC demonstrated the most consistent and strongest relationship (Figs 6A–6C, 6G, S3; Kendall rank correlation: τ_{B_IliacADC:MF Grade} = 0.44, τ_{B_IliacPDFF:MF Grade} = −0.30, τ_{B_IliacMTR:MF Grade} = 0.21; P < .05 for each). This relationship was independent of VOI by Kendall rank correlation (MF grade: VOI volume τ_B = −0.014, P = .91; iliac ADC: VOI volume τ_B = 0.023, P = .83). Correlations may be highest in the ilium because it is the site of bone marrow biopsy. On average, participants with MF demonstrated greater bone marrow cellularity than those with non-MF MPNs (Fig 6D, P = .03). Bone marrow cellularity demonstrated no correlation to fibrosis grade (Fig 6E).

Figure 6:

Bone marrow MRI correlates with MF fibrosis grade. (A) ADC, (B) PDFF (% fat), and (C) MTR (normalized ratio from 0 to 1) values from the ilium increase with bone marrow MF grade assessed with biopsy (Tukey multiple comparisons test, *P < .05, **P < .01, ***P < .001). (D) Bone marrow cellularity is greater in participants with MF in this cohort compared with participants with non-MF MPN (P = .03, unpaired t test). (E) Cellularity does not trend with MF fibrotic grade. (F) Using Fisher exact test, we demonstrate significant association between early fibrosis (MF grades 0–1) and overt fibrosis (MF grades 2–3) and abnormal ADC MRI (P < .01 odds ratio = 5.81). (G) We evaluated Kendall rank correlation between each MRI metric and the MF grade and bone marrow cellularity. All listed correlations were significant (P < .05); we do not list nonsignificant. PDFF is given as percentage of fat. ADC is in units of × 10⁻⁶ mm²/sec. MTR is a normalized ratio on a scale of 0–1. ADC = apparent diffusion coefficient, MF = myelofibrosis, MPN = myeloproliferative neoplasm, MTR = magnetization transfer ratio, PDFF = proton density fat fraction.

As ADC demonstrated the most consistent correlation with the MF fibrosis grade determined through bone marrow biopsy, we stratified participants with MF into two groups based on their corresponding bone marrow fibrosis grade: early (MF, 0–1) and overt (MF, 2–3) fibrosis. We then performed LDA on vertebral body, iliac, and femoral head ADC values to identify an LDA threshold for ADC at each anatomic site that best distinguished between early and overt fibrosis (Fig S4). We then characterized participants as having either 0–1 or 2–3 ADC anatomy values above the LDA thresholds for each anatomic site. This approach revealed a correlation between ADC values falling above the LDA threshold and the corresponding bone marrow fibrosis grade (Figs 6F, S5; P = .01, odds ratio = 5.81, sensitivity = 55% [11 of 20], specificity = 83% [19 of 23]). A participant with 2–3 ADC values above the identified thresholds had 5.81 higher odds of showing overt bone marrow fibrosis (MF, 2–3) than early to no bone marrow fibrosis (MF, 0–1), independent of VOIs based on a logistic regression analysis adjusted for VOI (Table S5). Despite the correlation observed for MRI metrics among various anatomic sites, these data demonstrated the value of identifying bone marrow abnormalities at different sites. Iliac ADC and MTR showed moderate correlations with MF grade and bone marrow cellularity, respectively (Fig 6G).

Discussion

Biopsy and aspiration are currently the only clinical methods for analyzing bone marrow pathology, including the expansion of hematopoietic cells and fibrosis, in MF, MPNs, and other hematologic malignancies. However, these invasive procedures have inherent limitations, such as small sample volumes, nondiagnostic analyses, and the reluctance of patients to undergo repeated biopsies. Therefore, oncologists continue to rely on spleen size, as determined with physical examination and/or imaging, as indirect measures of disease status in MF. The focus on spleen volume, however, continues to hinder the development of therapies that improve patient outcomes, such as approved drugs that reduce spleen volume with minimal effects on restoring normal bone marrow or extending survival (33). As the clinical management of and drug development in MF shift to arresting or reversing bone marrow pathologies, there is a clear need for reliable and reproducible methods for quantifying macroscale bone marrow architecture in individuals with MF (10). We investigated quantitative bone marrow MRI as an approach to overcome limitations of bone marrow biopsy and spleen volume to assess disease status in MF. Increased ADC in the ilium showed the best correlation with grade of bone marrow fibrosis (τ_{B_IliacADC:MF Grade} = 0.44), and participants with increased ADC in two to three anatomic sites had 5.81 higher odds ratio for overt bone marrow fibrosis (MF, 2–3). The advantages of MRI, including the noninvasive analysis of large volumes of bone marrow at multiple anatomic sites; the quantification of multiple MRI parameters, including measures of cellularity and, potentially, fibrosis; repeat follow-up scans at intermediate times; and assessment of potential heterogeneity within a bone and among different bones, suggest quantitative bone marrow MRI as a potential solution to this need.

By initially using MRI to characterize the bone marrow of healthy controls and participants with non-MF MPNs and MF, we established that MPNs and MF show increased bone marrow ADC and MTR. Bone marrow PDFF is lower in MF, reflecting the progressive loss of normal bone marrow fat due to the dysregulated proliferation of hematopoietic cells (5). Consistent with prior studies, PDFF decreased from femoral heads to the ilium and vertebral bodies (10). Although this study included a small number of healthy volunteers, the resulting 95% CIs were consistent with available reports for previous test-retest repeatability studies of PDFF and ADC in lumber vertebrae (PDFF repeatability coefficient = 10% and ADC repeatability coefficient = 125 × 10⁻⁶ mm²/sec) but lower than ADC repeatability coefficient reported for the ilium and femoral heads (repeatability coefficient = [85–95] × 10⁻⁶ mm²/sec) (25–28). General correlations existed for each MRI parameter across the lumbar vertebrae, ilium, and femoral heads, emphasizing the merit of assessing bone marrow from various anatomic sites. Our data demonstrated that ADC values were best correlated with various MF grades as determined with iliac bone marrow biopsy. Stratifying participants based on their corresponding ADC LDA scores also effectively distinguished between MF grade 0–1 and 2–3 disease.

The ability to identify grade 2–3 MF is clinically relevant owing to greater mortality in patients with overt bone marrow fibrosis (32). We identified no preference for specific pairs of abnormal anatomic regions (vertebral bodies and ilium vs ilium and femoral heads) in the higher odds ratio for grade 2–3 MF. Data showing higher ADC values with greater fibrosis are consistent with a prior study that showed increased ADC in fibrotic compared with normal liver tissue (34). Furthermore, previous studies reported elevated ADC in hypercellular human bone marrow and increases in ADC and MTR along with decreases in PDFF in bone marrow in mouse models of MF (12,14,17). Variations in the bone marrow architecture of different bones imply that biopsies from the iliac crest may not detect the full extent and magnitude of bone marrow disease. Furthermore, we established that healthy controls have greater bone marrow fat content (PDFF) and lower ADC and MTR than individuals with MF and determined that classifying participants based on number of anatomic sites with abnormal ADCs could distinguish between no or minimal (MF grade 0–1) and overt (MF grade 2–3) bone marrow fibrosis, a key distinction for patient outcomes (32).

Diagnostic imaging in MF is typically restricted to measurements of spleen volume for clinical trials, with only a limited number of imaging studies having assessed bone marrow in MF; however, the largest of these studies included only 35 participants. Previous studies showed increased water content and loss of fat, as measured with T1, short τ inversion recovery, and PDFF MRI sequences; however, these studies relied predominantly on qualitative observations (10,35,36). Bone marrow also showed evidence of increased perfusion and vascular permeability based on MRI enhancement with gadolinium-diethylenetriaminepentaacetic acid. PET with fluorine 18 (¹⁸F) fluorodeoxyglucose or ¹⁸F fluorothymidine demonstrated reduced radiotracer uptake in bone marrow with progressive fibrosis, likely owing to the loss of total and proliferating hematopoietic cells, respectively, and greater reliance on extramedullary hematopoiesis (37,38). However, changes in the uptake of these radiotracers provide only indirect evidence for bone marrow fibrosis, with data reported on a semiquantitative scale, potentially limiting reproducibility in trials involving more than a single institution. PET with gallium 68–labeled fibroblast activation protein inhibitor showed accumulation above background levels in bone marrow only in participants with grade 2–3 MF, reflecting the accumulation of activated myofibroblasts (39). The ability to distinguish overt from no or minimal fibrosis using the aforementioned methods is comparable to our study, although MRI provides additional information about bone marrow composition and anatomy in a single imaging study. As patients with MF may undergo treatment and follow-up for years, MRI removes the risk of repeated studies involving ionizing radiation.

Mechanisms accounting for increased ADC with cellularity and fibrosis in bone marrow remain undetermined. Normal bone marrow in older adults consists predominantly of fat, which has a very low ADC value, while increased water content from the progressive replacement of bone marrow fat with hematopoietic cells increases the range of achievable diffusion values. Additionally, increased hematopoiesis from greater vascular perfusion and local, fibrosis-associated, edema-like increases in extracellular fluid may influence ADC values (40). Of note, abnormal trabeculae and bone marrow fibrosis, both of which occur in MF, are histologic features of bone marrow edema (41). Therefore, further studies in preclinical models are needed to determine the specific pathologic changes that contribute to higher ADC in the bone marrow of individuals with MF.

We acquired two b values (0 and 800 sec/mm²) for diffusion-weighted imaging, which can only be fit with a monoexponential decay model while recognizing that multiexponential decay could exist due to the large voxel dimensions of images relative to the scale of heterogeneity in the bone marrow microstructure. To probe multiexponential behavior, however, we would need to acquire additional b values at the cost of increased scan time. Moreover, multiexponential behavior observed with low b values (b value < 200 sec/mm²) is typically attributed to perfusion, which we assume is low in bone marrow, or with high b values (b value > 1000 sec/mm²), owing to tissue microstructure. Diffusion-weighted imaging of the bone marrow is prone to a low signal-to-noise ratio owing to the magnetic susceptibility effects of bone trabeculae and suppression of fat signal, which may be the dominant source of signal in bone marrow. Because high b values further reduce signal-to-noise ratio, we stayed within the intermediate b value range and applied a simple monoexponential model.

There were some limitations of this study worth mentioning. First, we enrolled participants with MF and other MPNs from a single tertiary care institution; however, because MF is a rare cancer, we enrolled participants regardless of their treatment status or disease duration, increasing the heterogeneity of the study population, which included a large cohort for MF with driver mutation (JAK2, CALR, and MPL) frequencies comparable to those of prior reports (42). While participants with JAK2 mutations dominated the cohort (67%, 40 of 60), the heterogeneous cross-sectional cohort precluded meaningful comparisons of different driver mutations. However, such a comparison would be more informative with a more uniform study cohort, which is challenging to accomplish in a single-center trial with a rare malignancy such as MF. Second, additional factors such as age and degree of fitness also impact bone marrow composition. While using linear discriminant analysis to distinguish among participants with MF grades 0–1 and 2–3 was appropriate based on the size of this study, the LDA thresholds are specific to this participant cohort. Future studies may overcome this limitation by evaluating if bone marrow ADC values fall outside the healthy ADC range established using a larger cohort of healthy controls. Finally, while this analysis emphasized the utility of quantitative MRI metrics in evaluating progressive bone marrow disease, future studies would benefit from confirming the normal ranges for each MRI parameter based on a larger cohort of healthy controls.

Although we do not anticipate that MRI will eliminate the need for bone marrow biopsy for the diagnosis of MF, the spatial and temporal distributions of the disease at MRI provide novel information about bone marrow composition and will potentially markedly reduce the need for bone marrow biopsy to determine disease status over time. Therefore, future studies will utilize bone marrow MRI to monitor response to therapy, allowing us to analyze changes in MRI metrics over time. We have shown that this approach improved the assessment of treatment response in other diseases and accounts for variations in imaging data observed among individuals with single-time-point measurements. Therefore, we expect that comparing the magnitude and direction of change in imaging data for each participant relative to our previous MRI study will account for observed variations in values evident among individuals with single-time-point measurements.

In conclusion, MRI has merit as a noninvasive method to quantify the relevant metrics of disease in the bone marrow of individuals with MF. The MRI sequences used in this research, cleared by the U.S. Food and Drug Administration, are available on scanners from all major manufacturers. This study sets the stage for integrating quantitative bone marrow MRI into multicenter clinical trials evaluating existing or potential new drugs for MF. Multiple groups have developed phantoms that standardize imaging readouts across scanners from different vendors and institutions (26,43,44), with similar phantoms available commercially. These phantoms can be used to ensure the consistency of quantitative data obtained at multiple institutions over time, which is essential for deployment in multicenter clinical trials. The ability to assess bone marrow status over time with quantitative MRI lends well to the reversal of fibrosis and restoration of normal bone marrow architecture, a major focus of drug development in MF. We envision that the MRI methods presented in this study will ultimately improve the assessment of bone marrow in evaluating disease status and treatment response in individuals with MF.