AIMIC: Deep Learning for Microscopic Image Classification.

Area of Science:

• Artificial Intelligence
• Microscopy Image Analysis
• Computational Pathology

Background:

• Deep learning excels in image analysis but requires programming expertise, limiting clinical user adoption.
• Novice users face significant barriers to implementing advanced deep learning for microscopic image classification.

Purpose of the Study:

• To develop a user-friendly, code-free software solution for deep learning-based microscopic image classification.
• To lower the accessibility barrier for clinical users in applying artificial intelligence to microscopy image analysis.

Main Methods:

• Developed AIMIC (artificial intelligence-based microscopy image classifier), an out-of-the-box software enabling code-free deep learning.
• Integrated state-of-the-art deep learning techniques and data preprocessing within the AIMIC platform.
• Evaluated built-in deep learning networks on four benchmark microscopy image datasets.

Main Results:

• The AIMIC platform facilitates a complete deep learning pipeline, from training to inference, without programming.
• ResNeXt-50-32×4d demonstrated superior performance with 96.83% average accuracy and 96.82% average F1-score.
• MobileNet-V2 provided a favorable trade-off between accuracy (95.72%) and computational cost (0.109s inference time per sample).

Conclusions:

• The AIMIC platform democratizes AI application in microscopy image analysis for non-programmers.
• ResNeXt-50-32×4d is recommended for high-accuracy microscopic image classification.
• MobileNet-V2 serves as an efficient alternative for resource-constrained environments.