Detailed process description
Alveolar Detection Learning process.
In the first step, a Machine Learning model was created using IPSDK Explorer Smart Segmentation module. This operation consists in manually marking certain areas that are part of walls and alveoli. Once the model is created from a small part of the image, it can be reapplied on the whole image or on other similar images.
Once the IPSDK SMART Segmentation Machine Learning model is created, the Algorithm automatically segments the alveoli and the walls. During this intermediate step, it is possible to count the number of alveoli as well as to perform a size distribution measurement.
Wall distance map
Next step is computing the inter-pulmonary wall distance map. This Algorithm uses the binary image of the wall mask. Based on the contours, it will propagate inside the tissue by marking incrementally pixels ( based on the distance to the closest border).
Au final, la carte de distance fournit la cartographie globale des épaisseurs.
Extracting the mid-lines between alveoli.
Next, only the pixels present on the peak lines are considered to construct the wall thickness distribution histogram. These lines represent the separation lines generated from Watershed algorithm.
Remark : It is important to note that the values on these peak lines represent the distance to the nearest edge. It is important to double these values to obtain the thicknesses.
Thickness histogram computation
Finally, the process uses the pixel values under the midlines to automatically calculate the wall thickness distribution histogram.
This computation is directly performed using Masked Histogram measurement 2D function which is available as standard in IPSDK library.