Abstract
Objective
In-phase and opposed-phase chemical shift imaging (CSI) is a useful technique for assessing skeletal lesions. This study determined the frequency of significant differences in measurements obtained from longitudinal (coronal or sagittal) sequences to those obtained from axial sequences.
Methods
Chemical shift imaging was undertaken in 96 consecutive patients referred from the Musculoskeletal Sarcoma and Spinal Oncology services for assessment of possible bone tumours as part of a standard tumour protocol, which included turbo spin echo and inversion recovery sequences. For spinal lesions, CSI was obtained in the sagittal and axial planes, while for all other sites, it was obtained in the coronal and axial planes.
Results
The study included 49 (51.0%) males and 47 (49.0%) females with mean age 42.4 years (range 2-91 years). In 4 cases, 2 individual lesions were assessed, making a total of 100 lesions. Based on typical imaging features (n = 57) or histology (n = 43), 22 lesions (22%) were classified as non-neoplastic, 44 (44%) as benign neoplasms, 6 (6%) as intermediate-grade neoplasms, and 28 (28%) as malignant neoplasms. A significant discrepancy, wherein a lesion was classified as fat-containing (% SI drop >20%-25%) in the longitudinal plane, while in the axial plane it was classified as fat-replacing (% SI drop <20%-25%), or vice versa, occurred in 9%-14% of cases. However, this discrepancy had no appreciable effect on overall diagnostic accuracy, which was calculated at 79% for the longitudinal plane and 75%-80% for the axial plane.
Conclusions
Significant differences in CSI measurements occur in 9%-14% of cases based on imaging plane, but with no significant effect on diagnostic accuracy.
Advances in knowledge
Radiologists should be aware that CSI measurements in different planes appear to have significant differences in up to 14% of lesions. However, diagnostic accuracy does not seem to be significantly affected.
Keywords: Dixon, chemical shift imaging, marrow
Introduction
The Dixon sequence is a well-established technique for the assessment of bone marrow lesions.1–7 A particular value of in-phase (IP) and opposed-phase (OP) chemical shift imaging (CSI) obtained at 1.5 or 3 T utilizing either short echo time T1-weighted1,2,4 or T2W fast spin echo Dixon sequences8 is the ability to differentiate between fat-containing marrow lesions, such as focal marrow oedema, focal nodular marrow hyperplasia (FNMH), and atypical haemangiomas and marrow-replacing lesions, such as benign or malignant bone tumours. The former demonstrate a signal intensity (SI) drop of >20% on the OP sequence at 1.5 T,1,2,4 or >25% at 3 T,6 while marrow-replacing lesions demonstrate an SI reduction of <20% at 1.5 T1,2,4 and <25% at 3 T.6
However, CSI can also be associated with some pitfalls that radiologists need to be aware of, such as anomalous increase in SI on the OP sequence for mineralized/sclerotic marrow lesions.9 The senior author anecdotally noted a significant difference in CSI measurements obtained from sagittal/coronal (longitudinal) sequences and axial sequences in patients being assessed for marrow lesions, such that in 1 plane the lesion would have been considered a fat-containing lesion (SI drop on OP > 20%-25%) while in the other plane it would have been considered to be a fat-replacing lesion (SI drop on OP < 20%-25%). The aim of this study was to determine the frequency of such a finding, and how it may affect the diagnosis of either a non-neoplastic or a neoplastic lesion based on CSI.
Methods
The local Research and Innovation Centre of The Institute of Orthopaedics approved the study under the Integrated Research Application System number 262826, with no requirement for informed patient consent.
Between March 2021 and July 2021, 96 patients were referred from the Musculoskeletal and Spinal Oncology Services for investigation of suspected bone tumours. In 4 cases, there were 2 separate marrow lesions, and therefore a total of 100 marrow lesions were assessed.
Patient demographics
Data collected included patient age/gender, lesion location, and final diagnosis based on either histological confirmation from needle biopsy or characteristic imaging appearances, following which lesions were classified as either fat-containing non-neoplastic, or marrow-replacing neoplastic benign or neoplastic malignant lesions.
Imaging protocol
In addition to the routine bone tumour protocol consisting of a combination of T1-weighted turbo spin echo (T1W TSE), T2-weighted fast spin echo, short tau inversion recovery, proton density weighted fast spin echo, and spectral adiabatic inversion recovery sequences, all patients underwent IP and OP CSI in the sagittal plane (spinal lesions) or the coronal plane (all other skeletal sites), as well as in the axial plane for all lesions. In 43 (43%) cases, this was obtained at 1.5 T (Magnetom Sola; Siemens, Germany) and in 57 (57%) cases at 3 T (Ingenia; Phillips, Holland), but in all patients this was a short TE T1W sequence. The typical imaging parameters for the CSI techniques are presented in Table 1.
Table 1.
Typical MRI parameters for DIXON bone tumour imaging in lumbar spine.
| Field strength | Dixon | Plane | FOV (mm) | TR (ms) | TE (ms) | Slice thickness | Time (min) |
|---|---|---|---|---|---|---|---|
| 1.5 T | T1W (Vibe) | Longitudinal (sagittal 1 block) | 320×118.8 | 10 | TE1 = 2.2 TE2 = 4.4 | 3D | 3 min 30 s |
| Longitudinal (coronal 1 block) | 320×118.8 | 10 | TE1 = 2.2 TE2 = 4.4 | 3D | 3 min 30 s | ||
| Axial (1 block) | 323×59.4 | 10 | TE1 = 2.2 TE2 = 4.4 | 3D | 2 min | ||
| 3 T | T1W (mDixon) | Longitudinal (sagittal 1 block) | 350×90 | 17 | TE1 = 1.1 TE2 = 2.2 | 3D | 41 s |
| Longitudinal (coronal 1 block) | 350×90 | 17 | TE1 = 1.1 TE2 = 2.2 | 3D | 41 s | ||
| Axial (1 block) | 350×90 | 17 | TE1 = 1.1 TE2 = 2.2 | 3D | 39 s |
Abbreviations: FOV= field of view; TR = time to repeat; TE = time to echo; T1W = T1-weighted.
Imaging analysis
From the resulting IP and OP images, a region of interest (ROI) was drawn within the lesion on both the longitudinal and the axial sequences ensuring that the ROI corresponded to the centre of the lesion in both planes, and mean SI measurements were calculated using the PACS software. The percentage SI drop on the OP sequence was calculated as follows:
Lesions demonstrating >20% SI drop on the OP sequence at 1.5 T or >25% at 3 T were classified as fat containing/non-neoplastic, while those demonstrating <20% or 25% at 1.5 and 3 T, respectively, were classified as fat replacing/neoplastic. A Consultant Musculoskeletal Radiologist with 8 years of experience in bone tumour imaging (Reader 1) and a Consultant Musculoskeletal Radiologist with 27 years of experience in bone tumour imaging (Reader 2) independently recorded the IP and OP measurements for each case.
Statistical assessment
Inter-reader agreement for categorical variables was assessed using the kappa method, while agreement for continuous variables was assessed using the intra-class correlation.
An additional set of analyses quantified the number and percentage of patients in which there was a disagreement between longitudinal and axial sequences. A corresponding confidence interval for the percentage of disagreements was also calculated. Where a disagreement was observed, the nature of the disagreement was quantified, for example, longitudinal CSI indicated a non-neoplastic lesion, while axial CSI indicated a neoplastic lesion or vice versa.
In addition, a Bland-Altman plot (difference plot) was used to analyse the difference of percentage SI drop measurements between the 2 different orthogonal sequences.
Finally, the diagnosis of non-neoplastic or neoplastic lesions, based on longitudinal CSI and axial CSI for each lesion, was compared with the final diagnosis based on either histology or typical imaging features, allowing assessment of sensitivity, specificity, and overall accuracy of CSI for the 2 different imaging planes.
Results
The study included 96 patients, 49 (51.0%) males and 47 (49.0%) females with mean age 42.4 years (range 2-91 years). Lesion locations were as follows: acetabulum (n = 6), clavicle (n = 3), femur (n = 32), humerus (n = 9), ilium (n = 10), sacrum (n = 9), tibia (n = 9), vertebra (n = 15), radius (n = 1), rib (n = 1), fibula (n = 2), sternum (n = 1), pubis (n = 1), and scapula (n = 1). Based on typical imaging features (n = 57) or histology (n = 43), 22 lesions (22%) were classified as non-neoplastic, 44 (44%) as benign neoplasms, 6 (6%) as intermediate-grade neoplasms, and 28 (28%) as malignant neoplasms of which 10 were metastatic. Final diagnosis is presented in Tables 2 and 3. Inter-reader agreement for the longitudinal and axial CSI measurements are presented in Table 4, which shows a very high level of agreement for the individual CSI measurements and lesion classification into non-neoplastic or neoplastic based on % SI drop on the OP sequence. However, there was only moderate agreement for discrepancy between the lesion classifications based on longitudinal sequences compared to axial sequence. Table 5 shows details of disagreement and agreement between longitudinal and axial CSI for both Readers. The results suggested that for Reader 1 there was a disagreement between longitudinal and axial lesion classification in 9% of all patients (3 of 43 at 1.5 T and 6 of 57 at 3 T), the equivalent figure being slightly higher at 14% for Reader 2 (5 of 43 at 1.5 T and 9 of 57 at 3 T) (Figure 1). Where there was a disagreement, for Reader 1, there were a roughly equal number of patients with a disagreement in both directions. However, for Reader 2, where there was a disagreement, in almost 80% of cases the disagreement was such that the lesion was classified as neoplastic for the longitudinal sequence and non-neoplastic for the axial sequence. Nineteen out of the 100 longitudinal CSI cases were imaged in the sagittal plane. For Reader 2, there was no statistical significance between disagreements (3 of 14; 27.3%) compared with agreements (16 of 86; 22.9%) for sagittal plane vs axial plane CSI. Table 6 shows the diagnostic performance for the longitudinal CSI measurement and the axial CSI measurement using final diagnosis from imaging/histology as the gold standard. All measurements showed a sensitivity of ∼85%, with lower specificity values generally around 60%. Overall accuracy was between 75% and 80%. For Reader 2, in the 14 cases of disagreement, 3 had histopathological diagnosis confirmation with correct categorization in 2 of the 3 cases from coronal longitudinal rather than axial plane CSI measurements. Eleven of the disagreement cases had diagnoses made on imaging features alone, including stable imaging follow-up in 8 of the 11 cases. The other 3 cases had classical benign neoplastic imaging features such as enchondroma. Eight of the 11 cases had neoplastic (benign) diagnoses including 2 non-ossifying fibromas, benign notochordal cell tumour, fibrous dysplasia, intraosseous dysplasia, haemangioma, and enchondroma for which the longitudinal plane was correct in making the neoplastic diagnosis compared to axial plane. Only 1 of the 8 benign neoplastic cases was subcategorized correctly by the axial plane measurements. Of the 3 non-neoplastic lesions, all had stable imaging follow-up at 6 months, with axial rather than longitudinal plane suggesting correct categorization in 2 of the 3 cases. Therefore, in total, where there were disagreements between the planes of imaging, longitudinal measurements in 10 of 14 cases (71.4%) correlated more accurately with the offered radiological and histopathological diagnoses. A Bland-Altman plot demonstrated a negative bias of 2.6 for measurements of percentage SI drop for longitudinal vs axial CSI for Reader 1 (Figure 2). The confidence level between the 2 sequences included 0, suggesting no statistically significant difference between the results. There were 6 outliers beyond the levels of agreement in the study, of which 3 resulted in discrepant lesion classification. These outliers were equally split between 1.5 and 3 T.
Table 2.
Final diagnoses based on clinical and imaging findings.
| Non-neoplastic | Benign neoplastic | Intermediate neoplastic | Malignant neoplastic |
|---|---|---|---|
| FNMH (n = 9) | Fibrous dysplasia (n = 7) | Atypical cartilaginous tumour (n = 1) | Metastasis (n = 2) |
| Indeterminate (n = 3) | Enchondroma (n = 6) | Myeloma (n = 1) | |
| CRMO (n = 2) | Haemangioma (n = 5) | ||
| Marrow infarct (n = 2) | SBC (n = 4) | ||
| Insufficiency fracture (n = 1) | Enostosis (n = 3) | ||
| Marrow oedema (n = 1) | Intraosseous lipoma (n = 3) | ||
| NOF (n = 2) | |||
| BNCT (n = 2) | |||
| Atypical haemangioma (n = 1) | |||
| LSMFT (n = 1) |
Abbreviations: FNMH = focal nodular marrow hyperplasia; CRMO = chronic recurrent multifocal osteomyelitis; SBC = solitary bone cyst; NOF = non-ossifying fibroma; BNCT = benign notochordal cell tumour; LSMFT = liposclerosing myxofibrous tumour.
Table 3.
Final diagnoses based on histopathological results.
| Non-neoplastic | Benign neoplastic | Intermediate neoplastic | Malignant neoplastic |
|---|---|---|---|
| CRMO (n = 3) | ABC (n = 3) | GCT (n = 5) | Metastasis (n = 8) |
| Indeterminate (n = 1) | Enchondroma (n = 2) | Osteosarcoma (n = 5) | |
| Fibro-osseous lesion (n = 2) | Lymphoma B-cell (n = 3) | ||
| Mesench hypophos tumour (n = 1) | Chondrosarcoma (n = 2) | ||
| Haemangioma (n = 1) | Myeloma (n = 1) | ||
| Plasmacytoma-treated (n = 1) | |||
| Ewing sarcoma (n = 1) | |||
| Spindle cell sarcoma (n = 1) | |||
| Fibrosarcoma (n = 1) | |||
| Liposarcoma (n = 1) | |||
| Chordoma (n = 1) | |||
| LCH (n = 1) |
Abbreviations: CRMO = chronic recurrent multifocal osteomyelitis; ABC = aneurysmal bone cyst; GCT = giant cell tumour; LCH = Langerhans cell histiocytosis.
Table 4.
Inter-reader agreement.
| Outcome | ICC (95% CI) |
|---|---|
| Coronal/sagittal IPa | 0.98 (0.97, 0.99) |
| Coronal/sagittal OPa | 0.97 (0.95, 0.98) |
| Coronal/sagittal % SI drop | 0.94 (0.91, 0.96) |
| Axial IPa | 0.98 (0.97, 0.99) |
| Axial OPa | 0.97 (0.96, 0.98) |
| Axial % SI drop | 0.96 (0.94, 0.97) |
| Coronal/sagittal-axial % SI drop difference | 0.70 (0.58, 0.79) |
| Variable | Kappa (95% CI) |
| Coronal/sagittal lesion classification (NN or Neo) | 0.89 (0.70, 1.00) |
| Axial lesion classification (NN or Neo) | 0.88 (0.68, 1.00) |
| Coronal/sagittal-axial classification discrepancy (NN or Neo) | 0.56 (0.37, 0.75) |
Abbreviations: IP = in-phase; OP = opposed-phase; SI = signal intensity; ICC = interclass correlation; NN = non-neoplastic; Neo = neoplastic.
Analysis performed on log-transformed scale.
Table 5.
Disagreement between coronal/sagittal and axial CSI results.
| Variable | Reader | Category | Number (%) | 95% CI for % |
|---|---|---|---|---|
| Disagreement | 1 | No | 91 (91) | |
| Yes | 9 (9) | 4-16 | ||
| 2 | No | 86 (86) | ||
| Yes | 14 (14) | 8-22 | ||
| Nature of disagreementa | 1 | Cor/Sag NN, axial Neo | 5 (56) | |
| Cor/Sag Neo, axial NN | 4 (44) | |||
| 2 | Cor/Sag NN, axial Neo | 3 (21) | ||
| Cor/Sag Neo, axial NN | 11 (79) |
Abbreviations: NN = non-neoplastic; Neo = neoplastic; CSI = chemical shift imaging.
Figures only for lesions where a disagreement between methods is observed.
Figure 1.
A 27-year-old male who presented with non-specific pelvic pain. (A) Coronal T1-weighted turbo spin echo (T1W TSE) and (B) axial spectral adiabatic inversion recovery (SPAIR) MR images show an area of reduced T1-weighted (T1W) (arrow—A) and increased SPAIR (arrow—B) signal intensity (SI) in the posterior aspect of the right ilium. (C) Coronal in-phase (IP) and (D) opposed-phase (OP) chemical shift imaging (CSI) show an SI drop of 54% consistent with a non-neoplastic fat-containing lesion (arrows C and D). (E) Axial IP and (F) OP CSI show an SI drop of 16% consistent with a neoplastic marrow-replacing lesion (arrows E and F). The appearances were stable at 6 months of follow-up, and an imaging diagnosis of focal marrow hyperplasia was made.
Table 6.
Diagnostic performance of measurements in detecting lesion diagnosis.
| Measurement | Statistic | n/N | Estimate (%) (95% CI) |
|---|---|---|---|
| Longitudinal (Cor/Sag)—Reader 1 | Sensitivity | 66/79 | 84 (74, 91) |
| Specificity | 13/21 | 62 (38, 82) | |
| Accuracy | 79/100 | 79 (70, 87) | |
| Longitudinal (Cor/Sag)—Reader 2 | Sensitivity | 68/79 | 86 (76, 93) |
| Specificity | 11/21 | 52 (30, 74) | |
| Accuracy | 79/100 | 79 (70, 87) | |
| Axial—Reader 1 | Sensitivity | 67/79 | 85 (75, 92) |
| Specificity | 13/21 | 62 (38, 82) | |
| Accuracy | 80/100 | 80 (71, 87) | |
| Axial—Reader 2 | Sensitivity | 62/79 | 78 (68, 87) |
| Specificity | 13/21 | 62 (38, 82) | |
| Accuracy | 75/100 | 75 (65, 83) |
Figure 2.
Bland-Altman plot of agreement between % signal intensity (SI) drop on opposed-phase (OP) images for all lesions between longitudinal and axial plane imaging for Reader 1 with mean difference (orange dotted line), and upper (yellow dotted line) and lower levels of agreement (red dotted lines). This shows a negative 2.6 of longitudinal vs axial measurements, while confidence interval levels of the mean bias include the value 0, suggesting no statistical difference between the 2 measurements. There are only 6 outliers beyond the levels of agreement.
Discussion
This study investigated a previously unreported phenomenon of discrepancy of CSI measurements between longitudinal and axial imaging planes for skeletal marrow lesion, which occurred in 9%-14% of cases. This represents a potential further pitfall in the use of CSI for radiologists to be aware of when determining whether an indeterminate marrow lesion represents a fat-containing lesion, such as FNMH10 or an atypical haemangioma,11 or a benign/malignant marrow-replacing bone tumour.7,8 The reason for this discrepancy is unclear, but considering the high inter-observer agreement of the individual readings, it is considered a real phenomenon. It is unlikely due to partial voluming of normal marrow fat within the chosen ROI, since this was always obtained for all planes from the centre of the lesion and limited to the lesion. However, it did not affect the overall specificity, sensitivity, or diagnostic accuracy, which was very similar between the longitudinal and axial planes for both readers and to previously reported studies. The authors recognize that lower, only moderate, agreement between the readers for discrepancy in SI drop and resulting classifications compared with excellent inter-observer agreement for individual measurements could be attributed to recordings being close to the borderline <20/25% cut-off levels.
We are not aware that previous authors using CSI for assessment of marrow lesions have specified the need for obtaining measurements from longitudinal or axial planes. It would be most likely that measurements are taken in the plane that shows the lesion most clearly and its greatest volume, as has been our practice in the past. This usually equates to sagittal imaging in the spine and coronal imaging of the thorax, pelvis, and appendicular skeleton. This practice is supported, given that no significant difference was shown between sagittal vs coronal longitudinal plane for disagreements. Also, it may be prudent to consider the longitudinal over axial measurements if there is disagreement between them, given that longitudinal measurements in just over 70% correlated more accurately with the offered radiological and histopathological diagnoses, although it should be noted that disagreement occurred in a low percentage of cases.
This further pitfall of CSI for assessment of marrow lesions should be recognized by radiologists utilizing this technique, and is a further indicator that lesion classification between fat-containing marrow abnormalities, which are extremely unlikely to be malignant, and fat-replacing lesions should not be made solely by relying on % SI drop on OP Dixon CSI. With regard to suspected primary bone tumours, the importance of plain radiography for lesion characterization is again emphasized, particularly for assessment of growth rate of lytic lesions and the characterization of matrix mineralization.12,13
With regard to the role of MRI for assessment of focal marrow lesions, there are a wide variety of non-neoplastic marrow abnormalities, which can mimic bone tumours, either benign or malignant, and the importance of non-contrast enhanced T1W TSE sequences needs to be emphasized, with marrow-replacing lesions typically showing iso- to hypointensity compared to skeletal muscle.14
The study has several limitations. A definitive histological diagnosis was obtained in <50% of cases. However, it is well recognized that many benign primary bone tumours, such as enchondromas, atypical cartilaginous tumours, non-ossifying fibromas, fibrous dysplasia, and simple and aneurysmal bone cysts, have classical radiological features allowing a non-invasive imaging diagnosis. All imaging had been reviewed by a consultant musculoskeletal radiologist with over 27 years of experience in bone tumour imaging and all suspected malignant lesions and those which were considered indeterminate in nature were subjected to needle biopsy.
Apart from radiography, no other imaging techniques, such as diffusion-weighted imaging,15 were used. However, the aim of the study was not to determine the diagnostic accuracy of CSI for the assessment of suspected primary bone tumours, but to report a new potential pitfall of the technique.
In conclusion, we have reported a new potential pitfall in the use of CSI for the characterization of focal marrow lesions, this being a discrepancy between readings obtained in the longitudinal plane compared to the axial plane. This discrepancy was found in 9%-14% of cases, but had no appreciable impact on overall diagnostic accuracy.
Contributor Information
Asif Saifuddin, Department of Radiology, Royal National Orthopaedic Hospital, Stanmore, Middlesex HA7 4LP, United Kingdom.
Mohamed Ali, Department of Psychiatry, Leicestershire NHS Trust, Leicester LE3 9PQ, United Kingdom.
Rodney Santiago, Department of Radiology, Royal National Orthopaedic Hospital, Stanmore, Middlesex HA7 4LP, United Kingdom.
Ian Pressney, Department of Radiology, Royal National Orthopaedic Hospital, Stanmore, Middlesex HA7 4LP, United Kingdom; Institute of Orthopaedics and Musculoskeletal Science, University College London, Middlesex, HA7 4LP, United Kingdom.
Funding
None declared.
Conflicts of interest
None declared.
References
- 1. Disler DG, McCauley TR, Ratner LM, Kesack CD, Cooper JA.. In-phase and out-of-phase MR imaging of bone marrow: prediction of neoplasia based on the detection of coexistent fat and water. AJR Am J Roentgenol. 1997;169(5):1439-1447. [DOI] [PubMed] [Google Scholar]
- 2. Kohl CA, Chivers FS, Lorans R, Roberts CC, Kransdorf MJ.. Accuracy of chemical shift MR imaging in diagnosing indeterminate bone marrow lesions in the pelvis: review of a single institution’s experience. Skeletal Radiol. 2014;43(8):1079-1084. [DOI] [PubMed] [Google Scholar]
- 3. Dreizin D, Ahlawat S, Del Grande F, Fayad LM.. Gradient-echo in-phase and opposed-phase chemical shift imaging: role in evaluating bone marrow. Clin Radiol. 2014;69(6):648-657. [DOI] [PubMed] [Google Scholar]
- 4. Douis H, Davies AM, Jeys L, Sian P.. Chemical shift MRI can aid in the diagnosis of indeterminate skeletal lesions of the spine. Eur Radiol. 2016;26(4):932-940. [DOI] [PubMed] [Google Scholar]
- 5. Pezeshk P, Alian A, Chhabra A.. Role of chemical shift and Dixon based techniques in musculoskeletal MR imaging. Eur J Radiol. 2017;94:93-100. [DOI] [PubMed] [Google Scholar]
- 6. Kumar NM, Ahlawat S, Fayad LM.. Chemical shift imaging with in-phase and opposed-phase sequences at 3T: what is the optimal threshold, measurement method, and diagnostic accuracy for characterizing marrow signal abnormalities? Skeletal Radiol. 2018;47(12):1661-1671. [DOI] [PubMed] [Google Scholar]
- 7. van Vucht N, Santiago R, Pressney I, Saifuddin A.. Role of in-phase and out-of-phase chemical shift MRI in differentiation of non-neoplastic versus neoplastic benign and malignant marrow lesions. Br J Radiol. 2021;194(1119):20200710. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Saifuddin A, Shafiq H, Malhotra K, Santiago R, Pressney I.. Comparison of in-phase and opposed-phase T1W gradient echo and T2W fast spin echo Dixon chemical shift imaging for the assessment of non-neoplastic, benign neoplastic and malignant marrow lesions. Skeletal Radiol. 2021;50(6):1209-1218. [DOI] [PubMed] [Google Scholar]
- 9. van Vucht N, Santiago R, Pressney I, Saifuddin A.. Anomalous signal intensity increase on out-of-phase chemical shift imaging: a manifestation of marrow mineralisation? Skeletal Radiol. 2020;49(8):1269-1275. [DOI] [PubMed] [Google Scholar]
- 10. Rajakulasingam R, Saifuddin A.. Focal nodular marrow hyperplasia: imaging features of 53 cases. Br J Radiol. 2020;93(1112):20200206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Hoyle JM, Layfield LJ, Crim J.. The lipid-poor hemangioma: an investigation into the behavior of the "atypical" hemangioma. Skeletal Radiol. 2020;49(1):93-100. [DOI] [PubMed] [Google Scholar]
- 12. Costelloe CM, Madewell JE.. An approach to undiagnosed bone tumors. Semin Ultrasound CT MR. 2021;42(2):114-122. [DOI] [PubMed] [Google Scholar]
- 13. Benndorf M, Bamberg F, Jungmann PM.. The Lodwick classification for grading growth rate of lytic bone tumors: a decision tree approach. Skeletal Radiol. 2022;51(4):737-745. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Saifuddin A, Tyler P, Rajakulasingam R.. Imaging of bone marrow pitfalls with emphasis on MRI. Br J Radiol. 2022;96(1142):20220063. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Mansour TMM, El-Barody MM, Tammam H, Okasha A.. Role of diffusion-weighted MRI in differentiating between benign and malignant bone lesions: a prospective study. Clin Radiol. 2021;76(8):576-584. [DOI] [PubMed] [Google Scholar]