Segmentation Review

(***) Machine Learning Techniques for Biomedical Image Segmentation: An Overview of Technical Aspects and Introduction to State-of-Art Applications, Medical Physics, 2019, paper

Comparision between Traditional Methods and Deep Learning based Methods

Methods Advantage Disadvantage
Gredient based methods Fast Prone to image noise and artifacts that result in missing or diffuse organ boundaries.
Graph based methods (MRF et al.)   High computational cost due to iterative schedume.
Superivsed methods + prior knowledge capture shape well and generate more accurate results than unsupervised methods Limitated results when dealing with fuzzy boundaries
Deep learning methods Automatic feature extration scheme instead of manual feature extraction, More accurate Large computational cost, and large amount of data

Deep Learning based Methods

Methods Advantage Disadvantage
Patch based CNN Alleviate the coputational burden 1. loss of spatial infortation
2. too small patchs means less information could be extracted
3. may increase training time due to the duplicated computation of pixels
Challenges Current Solutions
Limited Training Data 1. Data Augmentation
2. Transfer Learning
2. Generate data using GAN
Volumetric 3D Data cost training time and computer memory 1. use 2D segmentation architectures instead of 3D architectures
2. use 2.5D architectures (an input data as several slice images, orthogonal images, maximum or minimum intensity projection)
Requires a lot of Hyperprameter Tuning None

(***) Understanding Deep Learning Techniques for Image Segmentation, ACM Computing Surveys, 2019, paper

Defination:

  • Image segmentation can be defined as a specific image processing technique which is used to divide an image into two or more meaningful regions.
  • Image segmentation can also be seen as a process of defining boundaries between separate semantic entities in an image.
  • From a more technical perspective, image segmentation is a process of assigning a label to each pixel in the image such that pixels with the same label are connected with respect to some visual or semantic property

Categories

Categories Task Notes
Semantic Segmentation Each pixel is classified into one of the predefined set of classes such that pixels belonging to the same class belongs to an unique semantic entity in the image. The semantics depends on the data and the problem that needs to be addressed
Saliency Detection focusing on the most important object in a scene  

Application

Comparization

  Traditional Methods Deep Learning Based Methods
Advantage   1. automated feature learning
Disadvantage 1. the results dependent on the quality of feature extracted by the domain experts
2. humans are bound to miss latent or abstract features for image segmentation		  

(***) Deep Semantic Segmentation of Natural and Medical Images: A Review, arXiv, 2019, paper

Categories:

Potentional Difficulty

difficulty solutions pros cons
medical image can be in high dimentions (2D and 3D) processed as sub-volumes (images) reduce computing cost prevent models capturing spatial information / relationships properly
lack of annotated data 1. semi- or un-supervised learning
2. encoding prior knowledge into model		    

Potential Future Directions

  • Going beyond pixel intensity-based scene understanding via incorporating prior knowledge.
  • Creating large 2D and 3D publicly available medical benchmark datasets for semantic image segmentation such as the Medical Segmentation Decathlon.
  • Exploring reinforcement learning approaches for semantic (medical) image segmentation to mimic the way human does delineation.