Skip to content

IMOP-lab/STPe-Pytorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

53 Commits
 
 
 
 
 
 
 
 

Repository files navigation

STPe-Pytorch:Stratified Trio-dimensional Pathway Encoding Framework

STPe: Stratified Trio-dimensional Pathway Encoding Framework

Minhong Sun, Han Yang, Tianyun Zhang, Jian Huang, Yihao Guo, Gaopeng Huang, Xuanbin Chen, Duanpo Wu, Shaowei Jiang, Xiaoshuai Zhang, Hong He, Xingru Huang, and Guan Gui

Hangzhou Dianzi University IMOP-lab

Methods

STPe

Figure 1: The HeON.overarching framework of the proposed Hybrid-distributed Elastic Ophthalmic Navigation System (HeON) for real-time macular surgery guidance.

We propose the HeON system, based on the IoMT, an efficient intraoperative stable delivery of quantization-aid 3D reconstruction system.

Next, we introduce our innovative STPe Segmention framework

Figure 2: The overall structure of proposed Stratified Trio-dimensional Pathway Encoding framework.

SpectroConv Module

Figure 3: SpectroConv Module.

The SpectroConv Module for extracting intricate time-frequency domain features and nuanced frequency domain information from image data.

HIT-EAFS Module

Figure 4: HIT-EAFS Module.

HIT-EAFS Module is designed to address the nuanced requirements of capturing both global and local feature representations effectively.

Experiment

Baselines

Training Results

Figure 5: Compare the results with other models

The proposed STPe segmentation results are compared with previous segmentation models, with evaluations across five categories: All (average of all categories), MH (Macular Hole), ME (Macular Edema), RA (Retina), and CR (Choroid). The best values for each metric are displayed in red, the second-best values in blue, and both the best and second-best metrics are indicated with highlights.

Figure 6: Visual segmentation performance

In the OIMHS dataset, visually compare STPe with other 14 segmentation models. We have selected 8 representative sequences for display. In these images, the segmentation results of macular holes are shown in red, macular edema in blue, the retina in yellow, and the choroid in green.

Ablation

Training Results

Figure 7: Ablation

The ablation of key modules in STPe was studied in the OIMHS data set. LDE (ResNet), GLE (MaxVit), and SCE (SpectroConv Discriminative Synthesis Encoder) corresponded to three branches in the Tri-dimensional Pathway Encoding, and NMCS, HIT-EAFS, and DFM corresponded to three key modules. The assessment included five categories: All, MH, ME, RA, and CR. The best values for each metric are displayed in red, the second-best values in blue, and both the best and second-best metrics are indicated with highlights.

Figure 8: Visual segmentation performance

Ablation experiments were executed based on the OIMHS dataset to evaluate the proposed STPe methodology. A1-A5 are denoted for group A and B1-B4 for group B, respectively. Four representative sequences were selected. In the depicted images, macular holes are delineated in red, macular edema in blue, the retina in yellow, and the choroid in green.

Reconstruction

Figure 9: Reconstruction

Example of two corresponding reconstruction outcomes (right) with original OCT images (left). The outcome delineates architectures of opaque retina, the third panel elucidates the transparent condition, highlighting the macular hole (red region) and macular edema (dark blue region).

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages