In the context of a development project for L’Oreal, the software application to automate the analysis of histological sections of skin had been developed in partnership with the Predictive Evaluations Department of L’Oreal Research and Innovation and the Mathematical Morphology Center of MINES ParisTech.
This software application is capable to:
- Load images of slides digitized by Hamamatsu NanoZoomer
- Detect the sections in the image of the whole slide
- Disable areas that should not be analyzed (automatically detected and/or user-manually defined)
- Automatically detect the layer interfaces (lights/stratum corneum/living epidermis/dermis)
- Let the user edit the detection
- Compute the morphological measurements of the layes/interfaces and quantify the melanin
- Generate an analysis report
Detection of the layer interfaces and the melanin
Classical image processing technics had been originally implemented to detect the layer interfaces (lights/stratum corneum, stratum corneum/living epidermis, and living epidermis/dermis). However, thanks to the emergence of the Deep Learning tehcnics, coupled with the classical image processing technics, it has been possible to obtain results comparable to those of experts but, as opposed to those of the experts, fully repeatable.
The Deep Learning technics also have the possibility to variability of the data (different types of skin and different types of staining) thanks to a relearning done by experts.
Graphical user interface of the software application for analyzing the histological sections of skin
The key strengths of this software application
- Capable to process huge images (Slides digitized at very high resolution)
- Highly reduced processing time
- Repeatable and not operator-dependent quantification thanks the automated detection of the interfaces and melanin
- Capable to process more samples to get more accurate statistical analysis
Specific Image Processing for Automated Analysis of Histological Cuts of Reconstructed In Vitro Skin and Ex Vivo Human Skin (en) 1 MB
Can Geodesic Information Help Convolutional Neural Networks? Application to the Segmentation of Fontana Masson Histological Images of Pigmented Reconstructed Epidermis (en) 417 KB