Medical imaging produces more data than almost any other branch of clinical medicine, and the number of trained eyes available to read it has not kept pace. Artificial intelligence (AI), and deep learning in particular, has moved from research laboratories into reading rooms over the past decade and now assists with tasks that range from sharpening noisy scans to flagging a possible cancer on a screening mammogram. This chapter introduces the subject for students and early-career professionals in radiology and the allied health sciences. It traces how the field grew from rule-based computer-aided detection to modern neural networks, explains the core ideas in plain language, and shows where AI enters the imaging workflow. Landmark clinical studies are summarised alongside the measures used to judge performance. The chapter also examines the practical limits of current tools, including bias, poor generalisation and thin clinical evidence, and outlines the ethical and regulatory questions that accompany their use.
Every medical image starts as a physical signal: an X-ray beam weakened by the chest, an echo returning from the liver, a faint radio wave from hydrogen nuclei, or a gamma ray leaving a tracer in bone. This chapter explains, in plain language, how the main imaging modalities generate and capture those signals and how the signals become digital pictures. It covers projection radiography, computed tomography, magnetic resonance imaging, ultrasound and nuclear medicine, and shows that all of them follow the same imaging chain. The second half deals with the digital image itself: pixels and voxels, sampling and matrix size, bit depth, windowing, and the DICOM format in which images are stored and shared. Spatial resolution, contrast, noise and artefacts are introduced as the measures of image quality, and typical radiation doses are compared. The chapter is written for students of radiography, imaging technology and the allied health sciences.
The plain radiograph is the commonest imaging examination in the world and the one most often read without a radiologist at hand, which has made it the busiest field for artificial intelligence in imaging. This chapter follows the radiograph through the department and describes where such tools are used. It begins at the X-ray unit, with positioning, exposure and automatic quality checks, and moves through image processing and bone suppression to the detection of disease. Chest radiography is treated in most detail, including nodule detection, the sorting of normal from abnormal films, worklist triage and tuberculosis screening, which matters greatly in India. Fracture detection, bone age and other musculoskeletal uses follow. The weaknesses that are particular to radiographs are set out, among them shortcuts learned from markers and portable equipment and unequal performance between patient groups. Simulations computed for the chapter illustrate the main ideas, and the role of the radiographer is discussed throughout.
Machine learning lets a computer improve at a task by studying examples instead of following hand-written rules, and deep learning is the branch of it that now dominates image analysis. This chapter sets out the foundations that a student of medical imaging needs before using or judging such systems. It begins with the main forms of learning and with the idea of describing an image by measurable features. Classical algorithms are then introduced, together with the central problem of overfitting. The chapter goes on to explain how an artificial neural network is built and trained, how convolutional networks read images, and which architectures suit classification, detection, segmentation and image synthesis. Practical matters follow: preparing and splitting data, augmentation, transfer learning, and the metrics used to measure performance. Worked examples computed from synthetic data illustrate each step, and the common mistakes that make published results unreliable are described.