Deep Learning based Phase Correction and Denoising for Accurate ADC Quantification

Motivation: Deep Learning based denoising methods improved image quality and scan efficiency of DWI imaging. However, artifacts and rectified noise floor limit their performance, leading to inconsistent improvement of ADC quantification. Goal(s): We aim to minimize artifacts and noise floor of DWI images and improve the accuracy and consistency of ADC quantification. Approach: Our approach is to combine DL-based phase correction and denoising methods to improve ADC quantification by minimizing artifacts and noise floor. Results: Compared to the current standard DL Denoising technique, we show improved ADC quantification in various challenging DWI applications. Impact: DLPC with DL denoising allows for a substantial improvement in ADC quantification compared to current standard DWI with and without DL Denoising technique. It will potentially allow wider adoption of ADC as a quantitative imaging biomarker in body Oncology.