Converting low-dose to higher dose 3D tomosynthesis images through machine-learning processes

摘要

A method and system for converting low-dose tomosynthesis projection images or reconstructed slices images with noise into higher quality, less noise, higher-dose-like tomosynthesis reconstructed slices, using of a trainable nonlinear regression (TNR) model with a patch-input-pixel-output scheme called a pixel-based TNR (PTNR). An image patch is extracted from an input raw projection views (images) of a breast acquired at a reduced x-ray radiation dose (lower-dose), and pixel values in the patch are entered into the PTNR as input. The output of the PTNR is a single pixel that corresponds to a center pixel of the input image patch. The PTNR is trained with matched pairs of raw projection views (images together with corresponding desired x-ray radiation dose raw projection views (images) (higher-dose). Through the training, the PTNR learns to convert low-dose raw projection images to high-dose-like raw projection images. Once trained, the trained PTNR does not require the higher-dose raw projection images anymore. When a new reduced x-ray radiation dose (low dose) raw projection images is entered, the trained PTNR outputs a pixel value similar to its desired pixel value, in other words, it outputs high-dose-like raw projection images where noise and artifacts due to low radiation dose are substantially reduced, i.e., a higher image quality. Then, from the “high-dose-like” projection views (images), “high-dose-like” 3D tomosynthesis slices are reconstructed by using a tomosynthesis reconstruction algorithm. With the “virtual high-dose” tomosynthesis reconstruction slices, the detectability of lesions and clinically important findings such as masses and microcalcifications can be improved.