Project Description

AI has shown great promise to improve image quality, quantiative accuracy and to solve undeterminsitic problems in medical imaging. Our group has applied generative adversarial networks (GAN) to denoise low dose static and gating myocaridal perfusion SPECT images, showing improved results as compared to conventional filterings and U-Net methods. In contrast to conventional deep learning methods based on training using other patients’ data, a “personalized” training scheme was also proposed for denoising with improved results. Besides image enhancement, image classifications for computer aided diagnosis are being investigated using different deep learning techniques.