Hu, Jingtong
(2020)
Achieve Realtime Object Detection for 4K and 8K Endoscopes with FPGA.
In: Pitt Momentum Fund 2020, University of Pittsburgh, Pittsburgh, Pa.
(Unpublished)
Abstract
"Endoscopes have been successfully and widely used to in varies diagnoses and procedures. The National Polyp Study showed that 70%–90% of Colorectal cancer are preventable with regular colonoscopies and removal of polyps. It is estimated that 85% of these “interval cancers” are due to missed polyps or incompletely removed polyps during colonoscopy. These misses come from both equipment factors and human errors. Most of existing colonoscopes are based on either 1080p HD image or 2K imaging technology. With such a resolution, images are blurred in many cases, which makes it impossible to distinguish diverse tissue architectures which appear similar to each other and are easily confused.
Recent development in UHD endoscopic system leads to much better image quality, which are likely to reduce miss diagnoses. With UHD endoscopes, another challenge to address to reduce miss diagnoses is the human factor. While doctors will gain more experiences and knowledge as they practice and learn, deep learning algorithms have shown great success. However, existing algorithms can only process 3-6 fps on 4K videos and 2fps on 8K videos, which are far away from the real-time requirement (25-30 fps) for endoscopy applications. The goal of this project is to develop a high performance accurate CNN accelerator for 4K and 8K videos to realize real time object detection and evaluate the performance with FPGA implementation."
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
Metrics
Monthly Views for the past 3 years
Plum Analytics
Altmetric.com
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}