[Paper Review] EfficientBioAI: Making Bioimaging AI Models Efficient in Energy, Latency and Representation

Yu Zhou, Justin Sonneck+5 more|arXiv (Cornell University)|Jun 9, 2023

Cell Image Analysis TechniquesBiochemistry, Genetics and Molecular Biology3 citations

TL;DR

EfficientBioAI is a plug-and-play toolbox that compresses bioimaging AI models using pruning and quantization to reduce energy consumption and inference latency by 30–80% and 2–5× speedup, respectively, without sacrificing accuracy—sometimes even improving it by reducing overfitting. The tool is compatible with common bioimage analysis frameworks like PyTorch, U-Net, and mmv_im2im, enabling efficient deployment on CPUs and GPUs across diverse biomedical imaging tasks.

ABSTRACT

Artificial intelligence (AI) has been widely used in bioimage image analysis nowadays, but the efficiency of AI models, like the energy consumption and latency is not ignorable due to the growing model size and complexity, as well as the fast-growing analysis needs in modern biomedical studies. Like we can compress large images for efficient storage and sharing, we can also compress the AI models for efficient applications and deployment. In this work, we present EfficientBioAI, a plug-and-play toolbox that can compress given bioimaging AI models for them to run with significantly reduced energy cost and inference time on both CPU and GPU, without compromise on accuracy. In some cases, the prediction accuracy could even increase after compression, since the compression procedure could remove redundant information in the model representation and therefore reduce over-fitting. From four different bioimage analysis applications, we observed around 2-5 times speed-up during inference and 30-80$\%$ saving in energy. Cutting the runtime of large scale bioimage analysis from days to hours or getting a two-minutes bioimaging AI model inference done in near real-time will open new doors for method development and biomedical discoveries. We hope our toolbox will facilitate resource-constrained bioimaging AI and accelerate large-scale AI-based quantitative biological studies in an eco-friendly way, as well as stimulate further research on the efficiency of bioimaging AI.

Motivation & Objective

• Address the growing energy and latency costs of increasingly complex bioimaging AI models in resource-constrained and large-scale biological studies.
• Overcome the lack of accessible, general-purpose model compression tools tailored for biologists and existing bioimaging AI frameworks.
• Enable efficient deployment of AI models on CPUs, edge devices, and diverse hardware backends without sacrificing predictive performance.
• Demonstrate that model compression can reduce overfitting by removing redundant representations, potentially improving accuracy.
• Facilitate eco-friendly, scalable AI-driven biological discovery by making efficient AI models widely accessible.

Proposed method

• Integrates neural network pruning and quantization as core compression techniques to reduce model size and computational load.
• Applies post-training quantization to convert 32-bit floating-point weights and activations to 8-bit integers, minimizing memory access and energy use.
• Employs structured pruning to remove low-importance filters or neurons, reducing model complexity while preserving feature representation.
• Supports plug-and-play integration with PyTorch-based models and common bioimaging tools like CellPose, DeepImageJ, and mmv_im2im.
• Chains multiple compression strategies (e.g., pruning followed by quantization) into customizable pipelines for optimal efficiency.
• Deploys compressed models via optimized inference engines across CPU and GPU platforms, ensuring cross-compatibility and low-latency inference.

Figure 1: overview of the toolbox. EfficientBioAI aims to compress the model and accelerate the bioimage tasks. a. Given a pretrained neural network (NN) in pytorch, the whole pipeline consists of two phases: compression (pruning, quantization) and inference. The compressed models are able to run on

Experimental results

Research questions

• RQ1 Can model compression via pruning and quantization significantly reduce energy consumption and inference latency in bioimaging AI models without degrading accuracy?
• RQ2 To what extent can compression techniques like quantization and pruning improve model generalization by reducing overfitting in low-data regimes?
• RQ3 How effective is the toolbox in accelerating large-scale 3D bioimage analysis tasks such as semantic segmentation of osteocytes in light-sheet microscopy?
• RQ4 Can the compressed models maintain fidelity in label-free fluorescence image prediction tasks, where subtle structural details are critical?
• RQ5 How compatible is the toolbox with existing, widely used bioimaging AI frameworks like mmv_im2im and U-Net, and what performance gains are achievable in practice?

Key findings

• The toolbox achieved 2–5× speedup in inference time across four different bioimage analysis tasks, including 3D segmentation and label-free imaging.
• Energy consumption was reduced by 30–80% during inference, significantly lowering the environmental footprint of AI-based analysis.
• In some cases, model accuracy improved after compression due to reduced overfitting, demonstrating the regularization effect of pruning and quantization.
• The compressed models maintained near-identical prediction quality to the original models, as validated by visual inspection and quantitative comparison with human-annotated ground truth.
• The toolbox successfully accelerated a two-minute inference task to near real-time, enabling faster iterative method development and large-scale screening.
• The approach demonstrated strong compatibility with popular bioimaging AI tools such as mmv_im2im, CellPose, and U-Net, enabling broad applicability across the field.

Figure 2: The effect of the sparsity ratio on different pruners, i.e., pruning algorithms (L1, L2, and FPGM) by observing the ratio of performance degradation (a.), energy saving (b.) and latency speedup (c.).