The study shows how deep learning can be used to detect cell image analysis.

By Brittney Grimes

Researchers have found a way to observe cell samples to study morphological changes — or the change in form and structure — of cells. This is significant because cells are the basic unit of life, the building blocks of living organisms, and researchers need to be able to observe what could influence the parameters of cells, such as size, shape, and density. 

Conventionally, cell samples were observed directly through microscopes by scientists to observe and discover any changes of the cells. They would look for morphological changes in the cell structures. However, they can now use artificial intelligence to make observations. Through using both computer science and a subset of artificial intelligence known as deep learning, researchers can now combine the methods to detect cell analysis. 

The study was published in the journal Intelligent Computing.

The importance of cell image analysis and using AI for research

Cell images have commonly been used in biomedical research and novel drug breakthroughs. The images show valuable information that encodes how cells react to external stimuli, or changes in environment, and intentional perturbations, or disturbances.

Researchers have used deep learning-based algorithms to automate the process of cell imaging, which is often done manually, and is a lengthy process.  The main objective of cell image analysis is to examine the phenotypic effects of different treatments and to discover the relationships between them. 

A breakthrough AI can track real-time cell changes revealing a key mystery in biology
Overview of deep learning-based cell image analysis.Deep Learning in Cell Image Analysis/Intelligent Computing 

The phenotypic outcomes refer to the observable characteristics within the cell structure. The study presented the three most critical tasks in cell image analysis, which are segmentation, tracking and classification.    

What are the three key tasks of segmentation, tracking and classification?

Segmentation is the fundamental principle for identifying, counting and morphological analysis of cell images. This key task is used to identify important features, which are divided into various parts, or segments, using deep learning. 

Classification acts as a downstream analysis – operations, such as testing the hypothesis, a prediction or an explanatory analysis – for phenotypic screening and cell profiling. This allows researchers to create an image of cell function by distinguishing between cells.

Tracking is the monitoring of cell images. This task usually occurs after segmentation. Researchers look for specific characteristics of cells, including any morphological changes, that can show the health status of the organism being studied. Examples of tracking involve immune response, cancer cell spreading and wound healing after injury. 

The research team reviewed the progression of deep learning applied to each of the key tasks. “In contrast to traditional computer vision techniques, a deep neural network (DNN) can automatically produce more effective representations than handcrafted representations by learning from a large-scale dataset. In cell images, deep learning-based methods also show promising results in cell segmentation and tracking,” the authors of the research said. 

Challenges of using AI for cell image analysis

There were some challenges mentioned by the researchers regarding the use of deep learning for cell image processing and analysis. “Deep learning has demonstrated an incredible ability to perform cell image analysis. However, there remains a significant performance gap between deep-learning algorithms in academic research and practical applications,” the authors said. The biggest challenge has to do with data, particularly data quality, data quantity and data confidence, or reliability.

The first limiting factor is the expenditure of the dataset with deep learning. The researchers stated that constructing a large-scale version of the dataset is strenuous because knowledgeable experts would have to meticulously assign labels image by image.  

The next challenge in using deep learning to assess cell analysis would be associated with imbalanced labels, meaning that the annotation of cell image datasets would be highly dependent on the professional skills of humas. This could lead to label imbalance, or preference within the labeled images, causing the numbers of labeled images for different classes to be unbalanced. The study also mentioned, similarly, label noise, in which human error causes training images of cells to be assigned incorrectly. 

A third challenge mentioned is called uncertainty-aware cell image analysis. This could lead to a problem because AI artificial neural networks cannot detect novel phenotypes without a method to reflect the reliability of the results of classification. 

Future application of AI to cell image analysis 

In the future, the researchers want to use artificial intelligence, specifically deep learning, to discover advanced concepts and underlying principles behind the traits and characteristics of cells. “Such successful applications demonstrate the ability of DNNs to extract high-level features and shed light on the potential capability of using deep learning to reveal more sophisticated life laws behind cellular phenotypes,” the researchers stated in the study.

The team hopes to improve cell image analysis by incorporating artificial intelligence into their assessments of cell imaging.