DeepCell Kiosk: scaling deep learning-enabled cellular image analysis with Kubernetes.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in imaging

Background:

• Deep learning models are powerful for biological image analysis.
• Analyzing large-scale imaging datasets with deep learning presents significant computational challenges.
• Existing workflows often struggle with scalability and cost-effectiveness for big data.

Purpose of the Study:

• To introduce DeepCell Kiosk, a cloud-native software solution.
• To demonstrate the dynamic scalability of deep learning workflows for large imaging datasets.
• To showcase the cost-efficiency of automated image analysis using cloud infrastructure.

Main Methods:

• Developed cloud-native software, DeepCell Kiosk, for dynamic scaling of deep learning.
• Utilized a cloud computing environment to process a large dataset of biological images.
• Implemented automated cell nucleus identification using deep learning models.

Main Results:

• Successfully identified cell nuclei in 10^6 1-megapixel images.
• Achieved processing time of approximately 5.5 hours for the large dataset.
• Demonstrated cost-effectiveness, with analysis costing around US$250, and potential for under US$100.

Conclusions:

• DeepCell Kiosk offers a scalable and affordable solution for deep learning-based biological image analysis.
• The software effectively addresses the challenges of processing large imaging datasets.
• Cloud-native architecture enables efficient and cost-effective large-scale biological data analysis.