A neural network works with arrays of numbers. But those numbers originate from physical measurements which are then formed into an image. What makes a structure bright? What limits the detail we can resolve? Does the pattern we observe in the image reflect biology, or was it introduced by the imaging system?
Answering questions like these helps us understand the signal we are dealing with, choose appropriate preprocessing methods, distinguish biological variation from imaging artifacts, and understand why an AI model may perform differently across different scanners or different acquisition protocols.
In this module we will follow the path from imaging physics and acquisition, through digitization and quantization, to issues you will encounter while working with digital images. We start by explaining how biological properties generate measurable signals, then look at how those signals become digital data. Finally, we consider the digital files you are likely to encounter and how to interpret and prepare their contents for machine learning.
2.1 What is a biomedical image?
2.5 Nuclear imaging — PET and SPECT
2.6 MRI — Magnetic resonance imaging
2.8 Optical microscopy and label-free imaging
2.9 Fluorescence microscopy and beyond
2.10 Cytology, histopathology and whole-slide imaging
2.11 Spatial transcriptomics and spatial omics