Computer vision (CV) is a field of AI that enables computers to interpret and understand visual information — turning pixels into decisions. CV systems can classify objects, detect their locations, track movement across video, segment regions at the pixel level, and measure quantitative features from images.
In healthcare, CV is transforming radiology, pathology, dermatology, ophthalmology, and surgery. A single hospital can produce millions of imaging studies per year — far more than radiologists can review with full attention. Modern CV systems now match or exceed human-level performance on specific tasks like detecting diabetic retinopathy from retinal scans or classifying skin lesions from dermoscopy images.
Before a computer can “see” an image, the image must be represented as numbers. An image is a grid of pixels — each pixel stores a color or intensity value at coordinates (x, y). Resolution is the number of pixels (width × height), and medical images range widely: dermoscopy (3000×4000) to ultrasound (640×480).
Grayscale: Single intensity value per pixel (0–255 for 8-bit). 0 = black, 255 = white. Used in many medical images (X-rays, CT scans).
RGB: Three values per pixel (Red, Green, Blue), each 0–255. Shape in memory: (height, width, 3) for NumPy, (3, height, width) for PyTorch.
Medical imaging uses specialized formats beyond standard JPEGs and PNGs.
DICOM (Digital Imaging and Communications in Medicine)
The standard format for medical imaging worldwide
Contains both pixel data AND rich metadata
Metadata includes: patient info (ID, demographics), acquisition parameters (modality, equipment), organizational hierarchy (study, series, instance)
Used by CT, MRI, X-ray, ultrasound, PET, and more
Other Common Formats: PNG (lossless, good for sharing), JPEG (lossy, not ideal for diagnostics), TIFF (multi-layer, common in pathology), NIfTI (3D/4D neuroimaging volumes)
