Content-based Retrieval of 3 D Medical Images Y. Qian, X. Gao , M. Loomes, R. Comley, B. Barn School of Engineering and Information Sciences Middlesex University, UK R. Hui, Z. Tian Department of Neurosurgery, General Navy Hospital, P. R. China
Contents 1. Background 2. Methodology 3. Experiment Results 4. Conclusion and Future Work
1. Background
MIRAGE (Middlesex medical Image Repository with a CBIR Archivin. G Environment) Ø Aim: To develop a repository of medical images benefiting MSc and research students in the immediate term and serve a wider community in the long term in providing a rich supply of medical images for data mining, to complement MU current online e-learning system. Ø So far 100, 000 2 D images and 100 images in 3 D form. Ø http: //image. mdx. ac. uk/ JSIC Ø Innovation in the use of ICT for education and research. Ø http: //www. jisc. ac. uk/
Content-Based Image Retrieval (CBIR) CBIR can index an image using visual contents that an image is carrying, such as colour, texture, shape and location. e. g. Query by Example Image(QBE)
Proposed Framework for MRIAGE
GIFT Framework GIFT(GNU Image Finding Tool) is open framework for content-based image retrieval and is developed by University of Geneva. Ø Query by example and multiple query Ø Relevance Feedback Ø Distributed architecture (Client - Server) Demo:
Content-Based 3 D Brain Image Retrieval 2 D brain images ----- 3 D Brain Ø Shape-based Surface of a 3 D object(e. g. tumor) Ø Texture-based Inside of a 3 D object( e. g. textures representing tissue structure properties Aim: To develop a fast texture-based 3 D brain retrieval method
2. Methodology
Proposed framework for 3 D image retrieval
Pre-processing 1) Spatial Normalization---Statistical Parametric Mapping (SPM 5) Transform each individual brain into a standard brain template 2) Divide 3 D brain into 64 non-overlapping equally sized blocks
Extraction of Volumetric Textures 1) 3 D Grey Level Co-occurrence Matrices (3 D GLCM) 2) 3 D Wavelet Transform (3 D WT) 3) 3 D Gabor Transform (3 D GT) 4) 3 D Local Binary Pattern (3 D LBP)
1) 3 D Grey Level Co-occurrence Matrices (3 D GLCM) 3 D GLCM is two dimensional matrices of the joint probability of occurrence of a pair of gray values separated by a displacement d = (dx, dy, dz). Ø 52 Displacement vectors: 4 distance * 13 direction = 52 Ø 4 Haralick texture features: energy, entropy, contrast and homogeneity Ø Feature vector: 208 components (=4 (features) * 52 (matrices)).
2) 3 D Wavelet Transform (3 D WT) 3 D WT provides a spatial and frequency representation of a volumetric image. Ø 2 scales of 3 D WT Ø Mean and Standard deviation Ø Feature vector: 30 components (2 (features) +15 (sub-bands))
3) 3 D Gabor Transform (3 D GT) 3 D GT generates a set of 3 D Gabor filters Gabor Transform: Ø 144 Gabor filters 4 (F) *6(θ)*6(Φ) =144 Ø Mean and Standard deviation Ø Feature vector: 288 components (2 (features) +144(filters))
4) 3 D Local Binary Pattern (3 D LBP) Local binary pattern(LBP) is a set of binary code to define texture in a local neighbourhood. A histogram is then generated to calculate the occurrences of different binary patterns. Ø 59 binary patterns Ø Feature vector: 177 components (=59(patterns)*3(planes)
Similarity Measurement ØHistogram Intersection(3 D LBP) Ø Normalized Euclidean distance (3 D GLCM, 3 D WT, 3 D GT)
Lesion Detection Assume bilateral symmetry of a normal brain along its mid-plane
3. Experimental Results
Test Dataset Ø 100 MR brain images Ø Size: 256 44 Ø DICOM (Digital Imaging and Communications in Medicine) format Ø Collected from Neuro-imaging Centre at Beijing General Navy Hospital, China
Experimental Results ------ Lesion Detection
Experimental Results -------Retrieval
Experimental Results -------Query time
4. Conclusion and Future work
1) Conclusion: Ø Comparative results demonstrate that LBP outperforms four 3 D texture methods in terms of retrieval precision and processing speed. Ø The query time with VOI selection offers 4 times faster operation that without. 2) Future work: Ø Test on the larger dataset Ø Plug 3 D image retrieval into GIFT framework (MIRAGE 2011)
Thank You.
Скачать презентацию Content-based Retrieval of 3 D Medical Images Y
cc8145d556af50098a892751f2e61738.ppt