Image Processing and Computer Aided Diagnosis in Computed Tomography of the Breast

摘要

Dedicated breast CT imaging is a novel breast imaging modality, which may improve lesion detection with a comparable radiation dose to conventional mammography. However, due to its cone beam geometry, breast CT suffers from image degradation due to scatter radiation. Moreover, the breast CT images divide the dose of mammography among hundreds of projection views, resulting in considerable quantum noise. It is therefore desirable to reduce scatter and noise in the reconstructed breast volume without loss of spatial resolution. Several new image-processing techniques were developed based on the unique physical properties of this modality. Using images obtained on a full field digital mammography system, the Gaussian noise model demonstrated effective scatter removal. A PDE technique taking into the account the non- uniform distribution of the noise in the projection image after the preprocessing step provides excellent denoised data with sharp edges. Applying a denoising technique before reconstruction provides better images than after reconstruction. The preliminary ROC study showed that with a fixed size lesion in real anatomical backgrounds, PDE-denoised images had higher detectability, higher CNR and better qualitative appearance. These promising new techniques for noise and scatter compensation pave the way for future implementations of dedicated breast CT.