Tissue Characterization by Image Analysis for Diagnostic Purposes João Sanches Institute for Systems and Robotics Instituto Superior Técnico
Noise Different Biomedical Image Modalities present different types of noise, e.g., additive or multiplicative. Usually, the noise should be removed. CT MRI LSFCM US
Speckle Processed by José Seabra (Biomedical PhD student) Ultrasound images usually present low quality (low SNR) Images are corrupted by speckle noise (multiplicative)
Medical Information However, the speckle pattern contains relevant medical information, e.g., fatty Liver.
Goal Decompose the image in textural and anatomical/morphological components. Characterize the texture/“noise” (speckle) Associate the texture characteristics with the disease. Classification and Quantification to Detect (Diagnosis) and Quantify (Severity Assessment) the disease.
Decomposition Noisy Synthetic Image Noiseless Image Noise Field
Classification Noise estimation Denoising Textural features Intensity and anatomical features Ultrasound Image Tissue characterization and Diagnosis
Decomposition Examples
“Noise” Analysis for Diagnostic Purposes Two Cases Liver Steatosis Atherosclerotic Plaques (carotids and coronaries)
Steatosis Steatosis is mainly a textural abnormality of the hepatic parenchyma due to fat accumulation on the hepatic vesicles (genetic, alcohol and obesity) Today, the assessment is subjectively performed by visual inspection
Steatosis Comparison with histological data
Steatosis Characterization Intensity Decay (m) Texture (E v,E h )
Diagnosis Processed by Ricardo Ribeiro (Biomedical PhD student)
Atherosclerotic Plaques
3D Diagnosis Global and local analysis
IVUS
Collaboration with the Centre de Visió per Computador, Universitat Autònoma de Barcelona, Profª Petia Radeva
IVUS IVUS Image decomposition
IVUS Automatic classification GT
Present and Future Work New texture characterization and classification methods Aterosclerotic plaques: 3D Extension of the IVUS Liver steatosis quantification with additional information from laboratorial analysis data.