Analyzing Microscopic Images: New Open-Source Software Enhances AI Model Sustainability

Artificial intelligence (AI) has become an indispensable part of micro-data analysis. However, while AI models are becoming better and more complex, computational power and related energy consumption are also increasing.

Therefore, researchers from the Leibniz Institute for Analytical Sciences (ISAS) and Peking University have developed a free compression software that allows scientists to run existing bioimaging AI models faster and significantly reduce energy consumption.

Researchers demonstrated their user-friendly toolbox, called EfficientBioAI, in an article published in Nature Methods.

"High network latency, such as large images, leads to higher computational power and ultimately increases energy consumption," said Dr. Jianxu Chen, head of the AMBIOM-Analysis of Microscopic BIOMedical Images research group at ISAS.

A well-known technique finds new applications

To avoid high latency in image analysis, especially on devices with limited computational power, researchers use complex algorithms to compress AI models. This means reducing the computational load in the models while maintaining prediction accuracy.

"Model compression is a widely used technique in the field of digital image processing, which can make models lighter and more environmentally friendly," explained Chen.

Researchers combined various strategies to reduce memory consumption, accelerate model inference speed, and save energy.

"These techniques are still relatively unknown in the field of bioimaging. Therefore, we hope to develop a ready-to-use and simple solution to apply them to common AI tools in bioimaging," said Yu Zhou, the first author of the paper and a doctoral student at AMBIOM.

Energy savings of up to 81%

To test their new toolbox, researchers led by Chen tested their software on several real-life applications. Through different hardware and various bioimaging analysis tasks, the compression technology was able to significantly reduce latency and decrease energy consumption by 12.5% to 80.6%.

"Our tests show that EfficientBioAI can significantly improve the efficiency of neural networks in bioimaging without compromising model accuracy," summarized Chen.

No special knowledge required

The authors are eager to have as many scientists as possible use EfficientBioAI in biomedical research. Researchers can install the software and seamlessly integrate it into existing PyTorch libraries.

Therefore, for widely used models such as Cellpose, researchers can use the software without making any changes to the code. To support specific modification requests, the team also provides demonstrations and tutorials. With just a few lines of code changes, the toolbox can also be applied to custom AI models.

About EfficientBioAI

EfficientBioAI is a ready-to-use open-source compression software for AI models in the field of bioimaging. The plug-and-play toolbox remains simple for standard use but offers customizable features. These include adjustable compression levels and easy switching between central processing units (CPUs) and graphics processing units (GPUs).

Researchers have been developing the toolbox and are working to make it available for MacOS, in addition to Linux (Ubuntu 20.04, Debian 10), and Windows 10. Currently, the focus of the toolbox is to improve the inference efficiency of pre-trained models rather than the efficiency during the training phase.