🦿Biomedical Engineering II Unit 6 – Image Processing and Analysis

Fundamentals of Digital Images

• Digital images are composed of discrete picture elements called pixels arranged in a 2D grid
  • Each pixel represents a specific location and has an intensity value
  • Pixel values are typically stored as integers (0-255 for 8-bit images)
• Image resolution refers to the number of pixels in an image and determines the level of detail
  • Higher resolution images have more pixels and can capture finer details (4K, 8K)
  • Lower resolution images have fewer pixels and may appear pixelated or blurry
• Color images use multiple channels to represent different color components
  • RGB (Red, Green, Blue) is a common color space for digital images
  • Each pixel in an RGB image has three values corresponding to the intensity of each color channel
• Grayscale images have pixels with a single intensity value representing shades of gray
  • Grayscale images are often used in medical imaging (X-rays, CT scans)
• Bit depth refers to the number of bits used to represent each pixel's intensity
  • Higher bit depths allow for a greater range of intensity values and more precise representation
• Image file formats define how image data is stored and compressed
  • Common formats include JPEG, PNG, TIFF, and DICOM (medical imaging)

Image Acquisition and Preprocessing

• Image acquisition involves capturing or obtaining digital images from various sources
  • Medical imaging modalities (X-ray, CT, MRI, ultrasound) produce images of the human body
  • Microscopy techniques (brightfield, fluorescence) capture images at the cellular level
• Preprocessing steps are applied to improve image quality and prepare images for further analysis
• Image denoising techniques reduce noise and artifacts that may be present in the acquired images
  • Gaussian filtering and median filtering are common denoising methods
• Contrast enhancement adjusts the intensity range of an image to improve visibility of features
  • Histogram equalization redistributes pixel intensities to cover the full range
• Image registration aligns multiple images of the same subject taken at different times or from different modalities
  • Allows for comparison and fusion of information from multiple sources
• Preprocessing may also involve cropping, resizing, or normalizing images to a consistent format
• Color space conversion transforms images between different color representations (RGB, HSV, LAB)
  • Useful for extracting specific color information or applying color-based processing

Spatial Domain Techniques

• Spatial domain techniques operate directly on the pixel values of an image
• Point operations modify each pixel independently based on its intensity value
  • Brightness adjustment adds or subtracts a constant value to each pixel
  • Contrast adjustment scales pixel intensities to expand or compress the intensity range
• Neighborhood operations consider the values of neighboring pixels when modifying a pixel
  • Convolution applies a kernel (small matrix) to each pixel and its neighbors to perform filtering or enhancement
  • Common convolution kernels include averaging, sharpening, and edge detection (Sobel, Prewitt)
• Morphological operations are used for image segmentation and shape analysis
  • Erosion shrinks objects by removing pixels from their boundaries
  • Dilation expands objects by adding pixels to their boundaries
  • Opening (erosion followed by dilation) removes small objects and smooths object boundaries
  • Closing (dilation followed by erosion) fills small holes and gaps in objects
• Spatial domain techniques are computationally efficient and intuitive to apply
• However, they may be sensitive to noise and can introduce artifacts if not applied carefully

Frequency Domain Analysis

• Frequency domain analysis transforms an image from the spatial domain to the frequency domain
  • Fourier transform decomposes an image into its frequency components (sinusoidal waves)
  • Low frequencies represent smooth variations, while high frequencies represent sharp edges and details
• Frequency domain techniques allow for selective manipulation of specific frequency ranges
• Low-pass filtering attenuates high frequencies, resulting in image smoothing and noise reduction
  • Ideal low-pass filter removes all frequencies above a cutoff threshold
  • Gaussian low-pass filter applies a gradual attenuation based on a Gaussian function
• High-pass filtering attenuates low frequencies, enhancing edges and fine details
  • Ideal high-pass filter removes all frequencies below a cutoff threshold
  • Gaussian high-pass filter applies a gradual attenuation to low frequencies
• Band-pass and band-stop filters selectively retain or remove a specific range of frequencies
  • Useful for isolating or suppressing certain image features or patterns
• Frequency domain analysis provides insights into the spectral content of an image
  • Power spectrum shows the distribution of energy across different frequencies
• Frequency domain techniques are particularly effective for periodic noise removal and texture analysis
• However, transforming between spatial and frequency domains can be computationally intensive

Image Enhancement Methods

• Image enhancement methods aim to improve the visual quality and interpretability of images
• Contrast enhancement techniques increase the distinction between different intensity levels
  • Global contrast enhancement applies a single transformation to all pixels (histogram equalization)
  • Local contrast enhancement adapts the transformation based on local image regions (adaptive histogram equalization)
• Sharpening techniques emphasize edges and fine details in an image
  • Unsharp masking subtracts a blurred version of the image from the original to highlight edges
  • High-boost filtering amplifies high frequencies to enhance sharpness
• Noise reduction methods suppress unwanted noise while preserving important image features
  • Gaussian filtering reduces Gaussian noise by averaging neighboring pixels
  • Median filtering reduces salt-and-pepper noise by selecting the median value in a neighborhood
  • Anisotropic diffusion smooths homogeneous regions while preserving edges
• Color enhancement techniques improve the appearance and contrast of color images
  • Color balancing adjusts the intensities of color channels to correct color casts
  • Saturation adjustment increases or decreases the vividness of colors
• Image inpainting techniques fill in missing or corrupted regions of an image
  • Useful for removing unwanted objects or restoring damaged portions of an image
• Image enhancement methods are subjective and depend on the specific application and desired outcome
  • Different techniques may be combined or applied iteratively to achieve the desired result

Segmentation and Edge Detection

• Image segmentation partitions an image into distinct regions or objects based on specific criteria
  • Regions are typically homogeneous in terms of intensity, color, or texture
  • Segmentation is a crucial step in many medical image analysis tasks (tumor detection, organ delineation)
• Thresholding is a simple segmentation technique that separates pixels based on their intensity values
  • Global thresholding uses a single threshold value to segment the entire image
  • Adaptive thresholding varies the threshold based on local image characteristics
• Region growing starts from seed points and iteratively expands regions based on similarity criteria
  • Pixels are added to a region if they satisfy the similarity criteria (intensity, color)
  • Region growing can handle complex shapes but may be sensitive to noise and seed point selection
• Watershed segmentation treats an image as a topographic surface and segments it based on watershed lines
  • Intensity gradients are interpreted as elevation, and segmentation lines are drawn at local minima
  • Watershed segmentation can produce precise boundaries but may oversegment the image
• Edge detection identifies sharp changes in intensity that correspond to object boundaries
  • Gradient-based methods (Sobel, Prewitt) compute the intensity gradient and detect edges at high gradient magnitudes
  • Laplacian-based methods (Laplacian of Gaussian) detect edges at zero-crossings of the second derivative
  • Canny edge detection combines Gaussian smoothing, gradient computation, and hysteresis thresholding for robust edge detection
• Segmentation and edge detection are essential for extracting meaningful regions and boundaries from images
  • Results can be used for further analysis, measurements, or visualization

Feature Extraction and Analysis

• Feature extraction involves computing quantitative descriptors that characterize specific properties of an image or image regions
• Intensity-based features capture the distribution and statistics of pixel intensities
  • Mean, median, and standard deviation describe the central tendency and variability of intensities
  • Histogram features (skewness, kurtosis) characterize the shape of the intensity distribution
• Texture features describe the spatial arrangement and patterns of pixel intensities
  • Gray-level co-occurrence matrix (GLCM) captures the frequency of pixel pairs with specific intensities and spatial relationships
  • Haralick features (energy, contrast, correlation) are derived from the GLCM and quantify texture properties
  • Local binary patterns (LBP) encode local texture patterns by comparing each pixel to its neighbors
• Shape features describe the geometric properties of segmented regions or objects
  • Area, perimeter, and circularity quantify the size and compactness of a region
  • Moment invariants (Hu moments) are shape descriptors that are invariant to translation, rotation, and scale
• Keypoint features identify distinctive points in an image that are stable under transformations
  • Scale-invariant feature transform (SIFT) detects keypoints at different scales and assigns orientation-invariant descriptors
  • Speeded up robust features (SURF) is a faster alternative to SIFT that uses Haar wavelet responses
• Feature selection and dimensionality reduction techniques help identify the most informative features
  • Principal component analysis (PCA) projects features onto a lower-dimensional space while preserving maximum variance
  • Feature ranking methods (e.g., Fisher score) evaluate the discriminative power of individual features
• Extracted features can be used for various tasks such as classification, clustering, or retrieval
  • Machine learning algorithms can be trained on feature vectors to classify images or detect specific patterns

Medical Image Applications

• Medical imaging plays a vital role in the diagnosis, treatment planning, and monitoring of various diseases
• X-ray imaging uses ionizing radiation to visualize internal structures
  • Chest X-rays are used to assess the lungs, heart, and bones
  • Mammography is a specialized X-ray technique for detecting breast abnormalities
• Computed tomography (CT) produces cross-sectional images by combining multiple X-ray projections
  • CT scans provide detailed visualization of organs, bones, and soft tissues
  • Used for diagnosing tumors, fractures, and internal injuries
• Magnetic resonance imaging (MRI) uses strong magnetic fields and radio waves to generate images
  • MRI provides excellent soft tissue contrast without ionizing radiation
  • Used for neuroimaging, musculoskeletal imaging, and cancer detection
• Ultrasound imaging uses high-frequency sound waves to visualize internal structures in real-time
  • Commonly used for prenatal imaging, cardiac imaging, and abdominal imaging
  • Doppler ultrasound measures blood flow velocity and direction
• Nuclear medicine imaging techniques (PET, SPECT) use radioactive tracers to visualize metabolic and functional processes
  • Positron emission tomography (PET) detects the distribution of a radioactive tracer to assess metabolic activity
  • Single-photon emission computed tomography (SPECT) captures the distribution of a gamma-emitting tracer
• Medical image analysis techniques assist in the interpretation and quantification of medical images
  • Segmentation of anatomical structures (organs, tumors) for volume measurement and treatment planning
  • Registration of images from different modalities or time points for comparison and fusion
  • Computer-aided detection (CAD) systems highlight potential abnormalities for radiologists to review
  • Quantitative imaging biomarkers extract measurable features related to disease progression or treatment response
• Medical image analysis plays a crucial role in improving diagnostic accuracy, treatment efficacy, and patient outcomes
  • Advances in machine learning and artificial intelligence are driving the development of automated analysis tools