Image reconstruction is the software inside a scanner that turns raw measurements into a picture that a doctor can read. A scanner never takes a photograph. A CT scanner measures how much X-ray passes through the body along thousands of lines. An MRI scanner measures radio signals from the body. An ultrasound machine measures echoes. None of these is a picture until software calculates one.
Here is how it works in a CT scanner. An X-ray tube spins around the patient. A row of detectors on the opposite side measures the X-rays that get through. Each measurement tells the software how much the body blocked along one straight line. From about 1,000 angles in one turn, the software works backwards. It calculates how dense each tiny point in the slice must be to explain all the measurements together. Bone is dense, so it shows up white. Air in the lungs shows up black.
Consider a child's CT scan at a hospital in Lucknow. Children are more sensitive to radiation, so the hospital wants the lowest dose that still gives a clear picture. The tube spins once in half a second. In that time, the detectors make about 1,000 sets of measurements. The software calculates each slice as a grid of 512 × 512 points, about 2.6 lakh points per slice, and the scan has about 300 slices. At a low dose, fewer X-rays reach the detectors, so the first picture is grainy. A deep learning model then removes most of the grain. The model was trained on pairs of scans of the same body, one at a low dose and one at a full dose. So the hospital can use a much lower dose and still get a clear picture.
There are three main methods. The oldest, called filtered back projection, is fast, but its pictures are grainy at low doses. Iterative methods repeat the calculation several times, comparing each guess with the measurements and correcting it. Deep learning methods are the newest. They need a lot of computation, so scanners carry GPUs, processors that do many calculations at the same time.
MRI uses a different calculation. Its signals are recorded as frequencies, and a mathematical method called the Fourier transform turns them into a picture. AI reconstruction lets an MRI scanner record fewer signals, so each scan takes less time.
Reconstruction is part of medical imaging equipment. It is written by the scanner makers' own engineers, many of them in Bengaluru. A reconstruction model is tested carefully, because it must never add anything to the picture that is not in the body, or remove anything that is. The same mathematics is used in industrial CT scanning, to see inside metal parts. The finished slices go on to 3D medical imaging and to AI radiology.