Deep learning models have proven to be very promising tools for analyzing large numbers of images. Over the past ten years, they have thus been introduced into various environments, including research laboratories.
In the field of biology, deep learning models could potentially facilitate the quantitative analysis of microscopy images, allowing researchers to extract meaningful information from these images and interpret their observations. However, training models to do this can be very difficult, as it often requires extracting features (i.e. cell number, cell area, etc.) from images of microscopy and manual annotation of training data.
Researchers from the CERVO Brain Research Center, the Institute for Intelligence and Data and Laval University in Canada have recently developed an artificial neural network that could perform in-depth analyzes of microscopy images at using simpler image-level annotations. This model, called MICRA-Net (MICRoscopy Analysis Neural Network), was presented in an article published in Intelligence of natural machines.
“Manual feature extraction from images is a time-consuming and tedious task, especially in cases where it must be performed by a trained expert,” said Anthony Bilodeau, a Ph.D. student at Laval University who performed the study, told TechXplore. “Although deep learning (DL) models for feature extraction are available, they still require training with annotations, which are often difficult to obtain. Our model (MICRA-Net) relies on a task of simple classification, asking the question: is structure present in the region of the image you are looking at or not?
By addressing this simple question, the model developed by the Université Laval team can predict the presence or absence of a specific structure in images using simple binary annotations. This dramatically reduces the time needed to annotate images and simplifies the training process, while allowing the model to tackle multiple microscopy image analysis tasks simultaneously.
“The weakness in supervision of our model stems from the way MICRA-Net is trained,” Bilodeau said. “The annotations required to train MICRA-Net are simple binary (yes or no) classification labels, which are much easier to obtain than complex precise labels, such as outlines of the structure of interest.”
Unlike other existing deep learning tools for microscopy image analysis, MICRA-Net can tackle complex tasks, such as semantic segmentation and detection, but using annotations of much simpler binary images. It achieves this by extracting essential information about the structure of interest from the gradient-class (i.e., grad-CAM) enabled maps.
“Combining grad-CAMs from multiple layers of the network allows the model to highlight the structure of interest in the image and can be used to generate accurate segmentation masks or to locate objects,” Bilodeau explained. “MICRA-Net also achieves similar or better performance on complex image analysis tasks compared to established baselines formed using weak supervision (e.g., bounding box annotations, squiggles). “
During the first evaluations carried out by the Laval University team, MICRA-Net obtained remarkable results, surpassing most of the models with which it was compared. In the future, it could thus be used by research teams around the world to solve complex image analysis problems and discover crucial patterns in microscopy images.
“Although some image analysis tasks may benefit from large, accurately annotated publicly available datasets for pre-training (e.g. kernel segmentation), we believe that MICRA-Net should be considered for data for which no precise annotation is readily available or can be easily obtained,” Bilodeau added. “For future research, we plan to test MICRA-Net on other complex datasets and also improve performance by investigating how other approaches can be combined for feature extraction.”
Anthony Bilodeau et al, Microscopy analysis neural network to solve detection, enumeration and segmentation from image-level annotations, Intelligence of natural machines (2022). DOI: 10.1038/s42256-022-00472-w
© 2022 Science X Network
Quote: A Weakly Supervised Machine Learning Model for Extracting Features from Microscopy Images (May 16, 2022) Retrieved May 17, 2022 from https://techxplore.com/news/2022-05-weakly-machine-features-microscopy- images.html
This document is subject to copyright. Except for fair use for purposes of private study or research, no part may be reproduced without written permission. The content is provided for information only.