Healthcare Analytics: Signal Processing & Pattern Recognition

Healthcare Analytics: Signal Processing & Pattern RecognitionTrần Thế TruyềnCenter for Pattern Recognition and Data AnalyticsDeakin University, AustraliaEmail: truyen.tran@deakin.edu.auURL: truyen.vietlabs.comHUST, Dec 2013Disclaimer: I don’t work on medical imaging specifically but general imaging.2Signal Types1D: EEG, ECG, radar, accelerometer, vital signs, blood sugar levels, etc.2D: Mammograms3D: angiography, ultrasoundMultimodal signals3Image Enhancement4Histogram Equalization5SubtractionTwo aligned (registered) images can subtract each other to enhance certain changing property.6MR angiography subtractionFiltering Techniques7Convolution8Gaussian averaging9Median filter10Truncated median filter11Mode filter12Comparison of filters13Image Segmentation14Thresholding15Edge Detector16Pixel classification17K-means Clustering18Energy Models & Graph Theory: Graph Cuts19Image Modeling20Markov random fields{‘Sky’, ‘Water’, ‘Animal’, ‘Car’, ‘Tree’, ‘Building’, ‘Street’}{124}Image from LabelMeApplication: Noise Removal22Feature Extraction23Histogram24Gabor wavelet transform25Statistical FeaturesMeanStandard deviationEntropySkewnessKurtosisEnergyContrast.26More Advance FeaturesLocal Binary PatternsSIFTHOGGIST27TextureBag-of-Words28Dimensionality Reduction29Dim Reduction TechniquesSubsamplingWavelet CompressionPCA: Principal Component AnalysisLDA: Linear Discriminant AnalysisICA: Independent Component AnalysisRBM: Restricted Boltzmann Machine30Why Dim Reduction?Faster processingLess noiseLess overfittingDe-correlationHigher-order features31Subsampling32What is PCA?33PCAEigenfaces341D Time-Series35Image as Time-Series36Spectrograms37Time-Series ClassificationDistance-basedFeature-basedModel-based38Dynamic Time Warping39Hidden Markov Models40Recurrent Neural Networks41From numerical signals into symbolsTemporal abstraction (AI technique)SAX (data mining technique)42