Classification of analyzable metaphase images using transfer learning and fine tuning

Area of Science:

• Genetics and Genomics
• Computational Biology
• Medical Imaging

Background:

• Chromosomal disorders can lead to significant structural and functional abnormalities.
• Accurate detection of metaphase stages is critical for identifying chromosomal defects.
• Chromosome pairing and analysis rely on precise identification of individual chromosomes.

Purpose of the Study:

• To investigate the efficacy of deep learning models, specifically VGG16 and InceptionV3, for analyzing human metaphase images.
• To leverage transfer learning and fine-tuning techniques for improved chromosome classification.
• To assess the performance of these models in detecting analyzable metaphase candidates.

Main Methods:

• Utilized VGG16 and InceptionV3 deep learning models with weights pre-trained on the ImageNet dataset.
• Applied transfer learning and fine-tuning approaches for chromosome image classification.
• Evaluated model performance using metrics such as true positive ratio, F-measure, precision, and recall across varying training set ratios.

Main Results:

• Achieved a high true positive ratio of 99% (±0.9) for both VGG16 and InceptionV3 networks.
• Reported F-measure, precision, and recall values of 99% (±1.0) for both models.
• Demonstrated comparable performance to state-of-the-art methods, with accelerated training phases due to transfer learning.

Conclusions:

• Transfer learning and fine-tuning with VGG16 and InceptionV3 are effective for accurate chromosome classification from metaphase images.
• The proposed deep learning approach significantly aids in the detection of chromosomal disorders.
• This method offers a computationally efficient and high-performing alternative for cytogenetic analysis.