Refine
Has Fulltext
- yes (2)
Is part of the Bibliography
- yes (2)
Document Type
- Journal article (1)
- Doctoral Thesis (1)
Language
- English (2)
Keywords
- 177Lu (1)
- 2-point Dixon (1)
- Bone Marrow Dosimetry (1)
- Bone Quantification (1)
- DEQCT (1)
- Deep learning (1)
- Denoising (1)
- Fat Quantification (1)
- Internal Dosimetry (1)
- MRI (1)
Institute
Bone marrow dosimetry is a topic of high interest in molecular radiotherapy. Predicting the level of hematological toxicity is one of the most important goals of nuclear medicine radiation dosimetry. To achieve this, it is necessary to quantify the absorbed dose to the active bone marrow, thus aiming at administering the most efficient therapy with a minimum level of adverse effects in the patient. The anatomical complexity of trabecular bone and bone marrow leads to the need of applying non-nuclear medicine imaging methods for determining the spatial distribution of soft tissue, adipose tissue, and bone in spongiosa.
Therefore, the two objectives of this dissertation are: i) to apply magnetic resonance imaging (MRI) for quantification of the fat volume fraction, and ii) to validate a method based on dual-energy quantitative computed tomography (DEQCT) for quantification of the trabecular bone volume fraction.
In a first step, an MRI sequence (two-point Dixon) for fat-water separation was validated in a 3 Tesla system by quantifying the fat volume fraction in a phantom and the lumbar vertebrae of volunteers and comparing with magnetic resonance spectroscopy (MRS). After successful validation, the fat volume fraction was retrospectively measured in the five lumbar vertebrae of 44 patient images acquired in the clinical routine. The two-point Dixon showed a good quantification of the fat volume fraction in the phantom experiment (-9.8% maximum relative error with respect to the nominal values). In the volunteers, a non-significant difference between MRI and MRS was found for the quantification of the fat volume fraction in volumes-of-interest with similar dimensions and position in both quantification methodologies (MRI and MRS). In the study with patient data, the marrow conversion (red → yellow marrow) was found to be age-dependent, and slower in males (0.3% per year) than in females (0.5% per year). Also, considerable variability of the fat volume fraction in patients of similar ages and the same gender was observed.
These results enable the use of two-point Dixon MRI in the quantification of the fat volume fraction in the bone marrow. Additionally, the constant marrow conversion during adulthood suggests that a patient-specific approach should replace the assumption of a constant cellularity volume fraction of 0.7 (reference man) (1,2) as proposed by the International Commission on Radiological Protection (ICRP).
In a second step, a quantification method based on DEQCT was validated in two CT systems: i) a clinical CT integrated into a SPECT/CT and ii) a dual-source computed tomography (DSCT) system. The method was applied in two phantoms: the first was used to validate the DEQCT method by the quantification of the hydroxyapatite volume fraction in three vials of 50 ml each and three different hydroxyapatite concentrations (100 mg/cm3, 200 mg/cm3, 300 mg/cm3). The second phantom was the European spine phantom (ESP), an anthropomorphic spine phantom. It was used to quantify the bone mineral content (BMC) on the whole vertebra and the hydroxyapatite volume fraction (VFHA) in the spongiosa region of each vertebra of the phantom. Lastly, the BMC of lumbar vertebrae 1 (LV1) and 2 (LV2) was measured in a patient using DEQCT and dual-energy X-ray absorptiometry (DEXA). Furthermore, the hydroxyapatite volume fraction (VFHA) and the bone volume fraction (VFB) was calculated for both the whole vertebrae and the spongiosa region of LV1 and LV2.
The measured and nominal hydroxyapatite volume fraction in the vial phantom showed a good correlation (maximum relative error: 14.2%). The quantification of the BMC on the whole vertebra and the VFHA on the spongiosa region showed larger relative errors than in the validation phantom. The quantification of BMC on LV1 and LV2 showed relative errors between DEXA and DSCT equal to 7.6% (LV1) and -8.4% (LV2). Also, the values of the VFHA (mineral bone) were smaller than the VFB. This result is consistent with the bone composition (mineral bone plus organic material).
The DEQCT method enables the quantification of hydroxyapatite (mineral bone) and bone (mineral bone plus organic material) in a clinical setting. However, the method showed an overestimation of the quantified mineral bone volume fraction. This overestimation might be related to the lack of detailed information on the CT X-ray spectra and detector sensitivity. Also, the DEQCT method showed a dependency on the CT reconstruction kernel and the chemical description of the materials to be quantified.
Based on the results of this work, the feasibility for quantifying the fat volume fraction and the bone volume fraction in the spongiosa in a clinical setting has been demonstrated/proven. Furthermore, the differences in fat volume fraction in females and males, as well as the variability of the fat volume fraction in subjects of similar ages, questions the approximation of the cellularity volume fraction by only a single ICRP reference value in bone marrow dosimetry for molecular radiotherapy. Lastly, this study presents the first approach for non-invasive quantification of the bone volume fraction (mineral bone plus organic material) for improved bone marrow dosimetry.
Background
In recent years, a lot of effort has been put in the enhancement of medical imaging using artificial intelligence. However, limited patient data in combination with the unavailability of a ground truth often pose a challenge to a systematic validation of such methodologies. The goal of this work was to investigate a recently proposed method for an artificial intelligence-based generation of synthetic SPECT projections, for acceleration of the image acquisition process based on a large dataset of realistic SPECT simulations.
Methods
A database of 10,000 SPECT projection datasets of heterogeneous activity distributions of randomly placed random shapes was simulated for a clinical SPECT/CT system using the SIMIND Monte Carlo program. Synthetic projections at fixed angular increments from a set of input projections at evenly distributed angles were generated by different u-shaped convolutional neural networks (u-nets). These u-nets differed in noise realization used for the training data, number of input projections, projection angle increment, and number of training/validation datasets. Synthetic projections were generated for 500 test projection datasets for each u-net, and a quantitative analysis was performed using statistical hypothesis tests based on structural similarity index measure and normalized root-mean-squared error. Additional simulations with varying detector orbits were performed on a subset of the dataset to study the effect of the detector orbit on the performance of the methodology. For verification of the results, the u-nets were applied to Jaszczak and NEMA physical phantom data obtained on a clinical SPECT/CT system.
Results
No statistically significant differences were observed between u-nets trained with different noise realizations. In contrast, a statistically significant deterioration was found for training with a small subset (400 datasets) of the 10,000 simulated projection datasets in comparison with using a large subset (9500 datasets) for training. A good agreement between synthetic (i.e., u-net generated) and simulated projections before adding noise demonstrates a denoising effect. Finally, the physical phantom measurements show that our findings also apply for projections measured on a clinical SPECT/CT system.
Conclusion
Our study shows the large potential of u-nets for accelerating SPECT/CT imaging. In addition, our analysis numerically reveals a denoising effect when generating synthetic projections with a u-net. Clinically interesting, the methodology has proven robust against camera orbit deviations in a clinically realistic range. Lastly, we found that a small number of training samples (e.g., ~ 400 datasets) may not be sufficient for reliable generalization of the u-net.