PADS-Net: GAN-based radiomics using multi-task network of denoising and segmentation for ultrasonic diagnosis of Parkinson disease.

Journal: Computerized medical imaging and graphics : the official journal of the Computerized Medical Imaging Society
Published Date:

Abstract

Parkinson disease (PD) is a prevalent neurodegenerative disorder, and its accurate diagnosis is crucial for timely intervention. We propose the PArkinson disease Denoising and Segmentation Network (PADS-Net), to simultaneously denoise and segment transcranial ultrasound images of midbrain for accurate PD diagnosis. The PADS-Net is built upon generative adversarial networks and incorporates a multi-task deep learning framework aimed at optimizing the tasks of denoising and segmentation for ultrasound images. A composite loss function including the mean absolute error, the mean squared error and the Dice loss, is adopted in the PADS-Net to effectively capture image details. The PADS-Net also integrates radiomics techniques for PD diagnosis by exploiting high-throughput features from ultrasound images. A four-branch ensemble diagnostic model is designed by utilizing two "wings" of the butterfly-shaped midbrain regions on both ipsilateral and contralateral images to enhance the accuracy of PD diagnosis. Experimental results demonstrate that the PADS-Net not only reduced speckle noise, achieving the edge-to-noise ratio of 16.90, but also attained a Dice coefficient of 0.91 for midbrain segmentation. The PADS-Net finally achieved an area under the receiver operating characteristic curve as high as 0.87 for diagnosis of PD. Our PADS-Net excels in transcranial ultrasound image denoising and segmentation and offers a potential clinical solution to accurate PD assessment.

Authors

  • Yiwen Shen
    Department of Cardiovascular Medicine, Heart Failure Center, Ruijin Hospital Lu Wan Branch, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • Li Chen
    Department of Endocrinology and Metabolism, Qilu Hospital, Shandong University, Jinan, China.
  • Jieyi Liu
    The SMART (Smart Medicine and AI-based Radiology Technology) Lab, Shanghai Institute for Advanced Communication and Data Science, Shanghai University, Shanghai, China; Key Laboratory of Specialty Fiber Optics and Optical Access Networks, School of Communication and Information Engineering, Shanghai University, Shanghai, China.
  • Haobo Chen
    The SMART (Smart Medicine and AI-based Radiology Technology) Lab, Shanghai Institute for Advanced Communication and Data Science, Shanghai University, Shanghai, China; Key Laboratory of Specialty Fiber Optics and Optical Access Networks, School of Communication and Information Engineering, Shanghai University, Shanghai, China.
  • Changyan Wang
    The SMART (Smart Medicine and AI-based Radiology Technology) Lab, Shanghai Institute for Advanced Communication and Data Science, Shanghai University, Shanghai, China; Key Laboratory of Specialty Fiber Optics and Optical Access Networks, School of Communication and Information Engineering, Shanghai University, Shanghai, China.
  • Hong Ding
    Department of Ultrasound, Zhongshan Hospital, Fudan University, No. 180 Fenglin Road, Xuhui District, Shanghai, 200032, China. ding.hong@zs-hospital.sh.cn.
  • Qi Zhang
    Department of Gastroenterology, The Affiliated Hospital of Qingdao University, Qingdao, China.