Skip to content

iamhxxn2/OCT_Segementation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 

Repository files navigation

💡 Segmentation layer of retina and the choroid in OCT images

Purpose

Background, 망막, 맥락막을 포함하고 있는 11classes Segmentation

Requirements

  • Python 3.6.13
  • PyTorch 1.10.2

Data Curation

Dataset Distribution

  • 11 classes

    • 망막의 경우 총 12개의 layer로 구성
    • 막 구조의 ELM(External Limiting Membrane) 제외
    • MZ(Myoid Zone), IZ(Interdigitation Zone)의 경우 층이 매우 얇기 때문에 각각 주변층과 하나의 Class로 통합
    • Choroid layer추가
  • Dataset

    • 각 Volume에 대해서 총 21 Slices이 존재

      • Volume : 각 환자의 OD(right eye), OS(left eye)에 대한 OCT Scan Data List
      • 황반의 중심을 지나는 11 slice를 기준으로 +-3mm 영역을 관찰하기 위해 5 slice ~ 17 Slice로 총 13개의 B-scan Images이 저장
    • 현재 633 Volumes이 존재 >> Total 8229(633*13) Images 존재

Image Processing & Cropping

  1. Labeling 과정에서 맥락막과 배경 사이에 새로운 Layer가 잘 못 발생하여 수정 image

  2. Mask 이미지에 다른 Class로 Labeling 되어있는 부분과 검은색 Hole들 제거 image

  3. 각 Mask Image에서 Start, Finish index를 찾은 후 Image와 Mask를 Cropping image

DataSplit

  • 각 중증도 별로 Patient를 기준으로 8:1:1 비율로 Split

Study Design

Pilot Study

  • Model : Unet (pretrained model)
  • Augmentation(Image Augmentation Library, albumnetation 이용)
    • Resize, RandomScale, RandomHorizontalFlip, CLAHE
  • Epoch : 50
  • Optimizer : SGD
  • Loss Functioin : CE(Cross Entropy)
  • Batch Size : 8
  • Learning Rate : 0.001

Problems

  • 현재 Mask를 기준으로 원본 이미지를 Cropping하였는데, 여기서 기준점 X
    • 사전 연구에서는 황반을 중심으로 +-3mm 만큼에 해당하는 pixel 범위의 영역을 사용
      • 황반을 찾는 알고리즘 구현
        1. 원본 이미지에서 edge를 추출
        2. edge 이미지의 각 열에서 가장 낮은 행 index만 추출
        3. 이렇게 추출한 line에서 가장 낮은 행 index를 가지고 있는 Pixel의 row, col 찾기

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors