Measuring bronchial wall thickness in a lung section
Reactiv’IP has developed a specific Python script, based on the IPSDK library, to automatically measure the thickness of the wall of a bronchiole from a lung section acquired by optical microscopy based on fluorochromes. This global approach makes it possible to generate a complete histogram of the thickness while avoiding tedious manual work usually carried out by an operator.
Lung section, vessel wall detection and automatic thickness measurement.
This application begins by detecting the bronchial duct using a Machine Learning model previously created from a few images. Once the canal has been segmented, the algorithm calculates a distance map from the mask corresponding to the walls. The thickness histogram is then calculated using the distance values derived from the peak line of this distance map.
Histogram of canal wall thickness in microns
Detailed description of the process
Training for the channel detection
The first step was to create a Machine Learning model using the Smart Segmentation module in IPSDK Explorer. The operation simply consists in manually marking a few areas of the walls and channel. Once the model has been built from a small area of the image, it can be reapplied to the whole image or to other images of the same type.
Channel segmentation
Once the Machine Learning model has been created, the algorithm, supplemented by a few morphology operations, is able to automatically segment the central channel.
Detection of marked tissue
The next step is to build a second machine learning model to segment the bronchial wall tissues that appear darker due to the fluorochrome used.
Wall segmentation
The previously constructed model allows us to segment the desired wall, but also some tissues outside the area of interest. To retain only the tissues of interest, the algorithm slightly expands the channel mask, and in the end retains only the marked tissues in contact with this expanded mask.
Thickness histogram calculation
Finally, the process uses a distance map on the segmented tissue to calculate the histogram of wall thicknesses.
This calculation is performed directly using the Masked Histogram measurement 2d function, available as standard in the IPSDK library.