The delivery of chemotherapy with radiotherapy improves survival compared with radiotherapy alone for patients with cervical cancer, however, high rates of acute hematologic toxicity occur when combining both therapies. The sparing of the red bone marrow (RBM) may reduce hematologic toxicity.^1,2 In Radiation Therapy Planning (RTP) RBM is identified based on a combination of PET-CT and water-fat MRI.³ Water-fat MRI⁴ can discriminate RBM and yellow bone marrow (YBM) regions because RBM has considerably lower fat content than YBM (0.5 vs. 0.8). In this study we investigated whether it is feasible to automatically segment RBM based only on MRI. The resulting RBM regions were benchmarked with FDG-PET images.Eleven patients with cervical cancer enrolled in the study and signed the informed consent approved by the IRB. Each patient underwent a whole-body FDG-PET/CT and MRI Dixon scan covering pelvis (1.5T MR scanner, axial GRE, flip angle 80˚, TE1/TE2=2.9/3.9 ms, TR=290 ms, acq. voxel 2x2x3mm³). MRI Dixon⁵ scan resulted in two (fat and water) images. Rigid registration was used to match the PET/CT and MR scans. Two patients were excluded due to the strong motion induced artifacts in MR images. A tool to segment red bone marrow (RBM) from water and fat MR images was developed in MATLAB. The automatic RBM segmentation tool included four main steps depicted in Fig.1. Since no bone mask can be obtained from MR images, an operator visually checks the resulting RBM delineations superimposed on MR image and the false positive region-of-interests (ROIs) outside the pelvic bones are identified and deleted. For step (1), the fat fraction (FF) range corresponding to the RBM was found based on the PET-CT data of four patients. On the PET-CT scans, the regions within the bony anatomy were delineated as RBM if their (normalized to body weight) standardized uptake values (SUV) were higher than the mean SUV over the whole body. Then these RBM delineations were superimposed on FF images and the mean range of the FF values among four patients was calculated. The RBM segmentation tool was validated with the data of the remaining five patients. The mean SUV in the delineated RBM regions were calculated.The optimal FF range was found to be: 0.3-0.7. For all the patients the normalized histograms of the FF values in the imaged volume were plotted (Fig. 2). The histograms had similar shape: two side peaks (mainly water and fat containing voxels) and a plateau. The FF corresponding to the RBM were on the plateau between two peaks. The false positive ROIs (≤3 per patient) occurred in the bowel region and could be easily identified. The delineations for one example patient from the validation group, superimposed on MR and FDG-PET images, are shown in Fig. 3. The mean SUV in the delineated regions were 1.58 (range 1.37-1.71) well above the reported cut-off SUV for RBM of 1.1⁶.The found range of FF values corresponding to RBM was centered around the theoretical value of 0.5, i.e. 50% fat content in the RBM. The width of the range can be explained by the physiological deviations of fat content. Next to it, the FF values used in this study are not absolute fat concentrations. The consistency of the FF values distributions in the patients (Fig. 2) may be used as an indicator of the tool’s robustness. The RBM segmentation tool does not provide a 100% accuracy and not all the red bone marrow regions were delineated (black arrows, FDG-PET, Fig.3). Nevertheless, the largest RBM regions were determined that, in principle, is sufficient for the sparing of RBM in RTP. The amount of false positive ROIs was very limited and they were easily identified by the operator. Therefore, the tool demands a minimal operator involvement and can be effectively included in the RTP workflow. However, artifact-free MR Dixon scans are necessary. To eliminate the operator involvement at all, the RBM segmentation tool can be combined with a pseudo-CT^7,8. The tool will not introduce an extra MR scan in the MRI-only workflow⁹ because it uses the same MR Dixon scan as the pseudo-CT. The conventional CT-based RTP could also benefit from MRI-only RBM delineations. For an effective RTP with the RBM sparing, the RBM discrimination within the pelvic bones is a necessity³. Moreover, a successful MRI-only RBM delineation could potentially exclude an expensive planning PET-CT scan for some patients.A feasibility of semi-automatic delineations of red bone marrow in pelvis derived from MRI-only was demonstrated. A broader study with larger number of patients is on going.