BME Frontiers / 2021 / Article / Tab 3

Review Article

Recent Advancements in Optical Harmonic Generation Microscopy: Applications and Perspectives

Table 3

Comparison of image analysis/machine learning techniques. Adapted from Ref. [103].

Analysis toolProsCons

2D FFT(i) Boundaries not required
(ii) Integrated in a variety of image analysis tools (i.e., FIJI and Matlab)
(i) Global approach
(ii) Difficult to detect small alterations in tissues with random fiber alignment
Curvelets(i) Extracts collagen fiber in relation to defined boundaries; extremely useful in breast cancer studies(i) Needs tumor/cell boundary for accuracy; may not be readily translated to other diseases
GLCM(i) Simple intensity-based approach
(ii) Widely available in multiple image analysis platforms (i.e., FIJI and Matlab)
(i) Highly variable and dependent on collagen coverage
(ii) Low sensitivity and specificity in most applications to date
Textons(i) Filters independent of fiber and image characteristics(i) Requires large image library
(ii) Does not identify discrete features
(iii) Computationally intensive: making it difficult to integrate
3D-voxel(i) Translatable to multiple imaging modalities
(ii) Rapid and highly accurate
(iii) Voxel-wise information
(i) Requires fiber-like features
(ii) May not be well suited for thick tissues
Deep learning(i) Less tedious and more efficient
(ii) Independent of specified image characteristics
(iii) Implementation of transfer learning enables use of small data sets
(i) Can require large image library
(ii) Complex training of neural networks

Please note that the Table of Contents page is currently not displaying content published since June 7, 2021. We are working to fix this issue as soon as possible and apologize for the inconvenience.