Effectiveness of Create ML in microscopy image classifications: a simple and inexpensive deep learning pipeline for...

Area of Science:

• Cytogenetics
• Computational Biology
• Machine Learning Applications in Microscopy

Background:

• Manual chromosome analysis is labor-intensive and time-consuming.
• Automated microscopic image acquisition systems generate large datasets (over 1000 images).
• A need exists for simple, inexpensive systems to efficiently select relevant images containing mitotic cells.

Purpose of the Study:

• To develop and apply a deep learning (AI) based classification system for chromosomal images.
• To enable non-data scientists to easily build and utilize AI models for image analysis.
• To efficiently select images containing mitotic cells from large datasets.

Main Methods:

• Applied a deep learning artificial intelligence (AI) classification system using Create ML on a Macintosh computer.
• Trained AI models using chromosome images from various plant species.
• Tested the system's ability to classify images from unseen plant species, tissue sections, and tetrads.

Main Results:

• AI models successfully classified images containing mitotic cells from both trained and untrained plant species.
• The system demonstrated efficacy on diverse biological samples, including tissue sections and tetrads.
• The developed system is inexpensive and user-friendly for scientists.

Conclusions:

• The deep learning system provides an efficient and accessible method for chromosomal image analysis.
• The AI tool can be easily trained by scientists using their own samples, reducing reliance on specialized expertise.
• This approach has broad applicability beyond chromosome analysis to other biology-related image datasets.