Study of the impact of image processing on the detection of microcalcifications in clinical digital mammography images acquired with different radiation doses

Abstract

Digital mammography is currently the most appropriate exam for breast cancer screening worldwide. However, despite the numerous benefits achieved by the mammographic examination, patients are exposed to small doses of radiation during the image acquisition process. It is known that the radiation dose used in the exam directly influences the quality of the mammographic image and, consequently, the performance of radiologists. Thus, a decrease in radiation exposure increases the safety of the exam, minimizing the risk of radio-induced tumors, but, as a consequence, it also increases the noise perceived in the image, compromising the effectiveness of the exam. On the other hand, an increase in radiation doses improves image quality, but also increases the risks of inducing new cases of cancer. Thus, the objective of this research project is to develop a study to assess the clinical impact of using image processing methods, based on noise filtering, as a tool to improve image quality and reduce radiation doses in digital mammography. The proposal consists, in a first step, in the formation of a database of mammography images of retrospective examinations acquired with the standard radiation dose. A computational method for inserting clusters of microcalcifications should be used to control the location, size and contrast of the lesions in each image. To simulate acquisitions with different radiation doses, a method of quantum noise injection, developed by our team in previous work, will be used. Clinical imagens at different radiation doses will be processed by restoration methods specifically developed for filtering quantum noise in digital mammography, in order to ensure that they have equivalent or superior quality to the images acquired with the standard radiation dose. The restoration methods that will be used should be based on mathematical models of noise and also on artificial neural networks. Finally, a set of clinical trials will be carried out to assess how different radiation dose reduction rates and image processing methods affect the performance of radiologists, both in localizing and detecting breast microcalcifications of clinical interest. The project is novel and will be carried out in partnership with the Barretos Cancer Hospital (Pio XII Foundation - Brazil), Ribeiro Preto Medical School (FMRP/USP - Brazil), Tampere University (TUNI - Finland), Rensselaer Polytechnic Institute (RPI - USA) and with the University of Pennsylvania (Penn - USA). As a result of this study, it is expected that the risks of breast cancer induction from radiation exposures will be minimized, and image quality will be improved, increasing diagnostic efficiency of digital mammography, which will benefit breast cancer screening programs and the female population health. (AU)