A deep learning model to predict breast implant texture types using ultrasonography images: feasibility development study

HIGHLIGHTS

• What: Objective: objective of this study is to determine feasibility of using a deep learning model to classify textured type of and make robust predictions from images from heterogeneous sources. From stratified 5-fold cross-validation, the model showed an average AUROC of 0.98 and PRAUC of 0.88 in the Canon dataset captured with the Canon ultrasonography device (D1) (Supplementary data 2). The study reveals that deep learning models may be vulnerable to medical images from heterogeneous sources due to unseen distribution. This study showed uncertainty in the model`s predictions, with the mean distribution . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or Email

  Password

  Remember me

  Lost your password?

  If you don't have an account, you can create one here.