Logo of the Physikalisch-Technische Bundesanstalt
Into the Future with Metrology - The Challenges of Medical Technology

Digitalization in the Health Care System

Model-based observer and machine learning for high-quality images

Mammography
A procedure for quality assurance in mammography screening, which is based on machine learning, has been developed at PTB. Thereby, the contrast-detail curve – as a measure of the image quality – is determined for the first time on the basis of only one single image. The new method is robust and more precise than the methods used so far. (picture alliance / dpa)

PTB scientists in Braunschweig and Berlin are both comparing and developing modern methods for measuring the image quality of X-ray images, for example in computed tomography or mammography. These methods must provide a measure of the uncertainty of the determined image quality while being, at the same time, objective and economical. For this reason, so-called phantoms are used to represent the patient and model-based mathematical observers (model observers) are used as a substitute for radiologists. Image quality analysis with model-based observers is a rapidly growing field of research. For example, PTB uses machine learning (Deep Learning) in order to determine – in a particularly efficient way – the so-called contrast-detail curves which are required for acceptance and constancy tests on mammography scanners. The large image database required for the training of the neural network was generated with a mammography simulation program developed at PTB. Both procedures – the procedures that are based on machine learning and the "classical" model-based observers – are being developed with the aim of being able to use them for quality assurance and standardization in the foreseeable future. From the figures of merit for image quality and dose, a target value will ultimately be developed for the optimization of radiological diagnostics – with the aim of achieving the best possible image quality with a low dose.

 

Departments involved
Opens internal link in current window6.2 Dosimetry for Radiation Therapy and Diagnostic Radiology

Opens internal link in current window8.4 Mathematical Modelling and Data Analysis