Name
Abstract | ||
|---|---|---|
Multiple myeloma is a tumor disease in the bone marrow that affects the skeleton systemically, i.e. multiple lesions can occur in different sites in the skeleton. To quantify overall tumor mass for determining degree of disease and for analysis of therapy response, volumetry of all lesions is needed. Since the large amount of lesions in one patient impedes manual segmentation of all lesions, quantification of overall tumor volume is not possible until now. Therefore development of automatic lesion detection and segmentation methods is necessary. Since focal tumors in multiple myeloma show different characteristics in different modalities (changes in bone structure in CT images, hypointensity in T1 weighted MR images and hyperintensity in T2 weighted MR images), multimodal image analysis is necessary for the detection of focal tumors. In this paper a pattern recognition approach is presented that identifies focal lesions in lumbar vertebrae based on features from T1 and T2 weighted MR images. Image voxels within bone are classified using random forests based on plain intensities and intensity value derived features (maximum, minimum, mean, median) in a 5 x 5 neighborhood around a voxel from both T1 and T2 weighted MR images. A test data sample of lesions in 8 lumbar vertebrae from 4 multiple myeloma patients can be classified at an accuracy of 95 % (using a leave-one-patient-out test). The approach provides a reasonable delineation of the example lesions. This is an important step towards automatic tumor volume quantification in multiple myeloma. |
Year | DOI | Venue |
|---|---|---|
2015 | 10.1117/12.2081990 | Proceedings of SPIE |
Keywords | Field | DocType |
multiple myeloma,tumor segmentation,pattern recognition,multimodal image analysis | Voxel,Computer vision,Lesion,Feature (computer vision),Segmentation,Multiple myeloma,Lumbar vertebrae,Image segmentation,Artificial intelligence,Radiology,Hyperintensity,Physics | Conference |
Volume | ISSN | Citations |
9414 | 0277-786X | 0 |
PageRank | References | Authors |
0.34 | 1 | 3 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| andrea franzle | 1 | 0 | 0.34 |
| jens hillengass | 2 | 0 | 0.34 |
| Rolf Bendl | 3 | 45 | 11.54 |