Awesome
DOVE: Doodled Vessel Enhancement for Photoacoustic Angiography Super Resolution
🚀🚀🚀 We're thrilled to announce that our paper has been accepted for publication in Medical Image Analysis!
🚀 Brief
This project presents a PyQt implementation for Super-Resolution Photoacoustic Angiography Assisted by Images Forged from Hand-Drawn Graffiti. Leveraging the concept of Image Super-Resolution via Iterative Refinement Project, our application allows users to draw graffiti on a board, generating its corresponding photoacoustic version within minutes. The implementation is coded using PyQt5.
🚀 How to Use
# First, download the model from Google Drive and place it in ./experiments
# [Google Drive Model](https://drive.google.com/file/d/1XWXWG4DAw0ZPd0N_3r-7jQMF5WMKyOG0/view?usp=share_link)
cd dir_of_the_project
python ./main.py
🚀Schematic
<img src="./readme/fig000.png" alt="show" style="zoom:50%;" />(a)-(g): The schematic illustrates the training process for super-resolution blood vessel images in Photoacoustic Angiography (PAA) forged from hand-drawn doodles. It includes steps such as generating rain-like noise, hand-drawn doodles, input image generation, normalized Gaussian noise, PAA image generation, normalized PAA image, and reconstructed super-resolution PAA image. (a')-(f'): The corresponding schematic for training super-resolution blood vessel images in PAA forged from cropped images.
🚀GUI snapshot
Note: PyQt5-based hand-drawn graffiti for photoacoustic images generation. The GUI with graffiti and without graffiti is displayed below.
<img src="./readme/fig00.png" alt="show" style="zoom:50%;" /> <img src="./readme/fig01.png" alt="show" style="zoom:50%;" /> <img src="./readme/fig02.png" alt="show" style="zoom:50%;" />🚀Results
Forging realistic photoacoustic images.
Note: We generate different kinds of photoacoustic images by adding noise of Gaussian distribution. The left two are input images, while the right two are their corresponding photoacoustic versions.
<img src="./readme/fig1.png" alt="show" style="zoom:200%;" />The left two are input images while the right are their photoacoustic version.
* 🚀Photoacoustic image super-resolution
Note: The shallow and deep feature extraction of SwinIR enables us to utilize the self-similarity of blood vessel images. These results also prove the abilities of our proposed method in forging photoacoustic images.
<img src="./readme/fig2.png" alt="show" style="zoom:51%;" />(a) High-resolution ground truth; (b) Reconstructed by Bicubic algorithm; (c)-(f) Reconstructed by SwinIR trained with various datasets. (c) COCO dataset; (d) mouse brain PAA images; (e) real human lips PAA images; (f) forged human lips PAA images.
Key words
** Photoacoustic, Microscopy, Diffusion Model, Biomedical Imaging, Medical Imaging