Using Image Recognition to Process Unbalanced Data in Genetic Diseases From Biobanks

Area of Science:

• Genomics
• Biomedical Informatics
• Medical Imaging

Background:

• Human biobanks are crucial for precision medicine and genetic disease research.
• Increasing medical imaging data necessitates advanced image recognition in biomedicine.
• Case-control data imbalance is a common challenge in human biobanks, impacting genetic disease research.

Purpose of the Study:

• To address case-control data imbalance in the Taiwan Biobank for genetic disease research.
• To evaluate a novel method, TW-SMOTE, for balancing biobank data.
• To compare the efficacy of TW-SMOTE with existing methods like SAIGE and UK Biobank (UKB) data.

Main Methods:

• Utilized Manhattan plots and genetic disease information from the Taiwan Biobank.
• Applied Synthetic Minority Oversampling Technique (SMOTE) to create a balanced dataset, termed TW-SMOTE.
• Employed a deep learning image recognition system to analyze the TW-SMOTE dataset.

Main Results:

• TW-SMOTE effectively adjusted the case-control ratio imbalance in the Taiwan Biobank.
• The TW-SMOTE method achieved results comparable to the statistical method SAIGE.
• TW-SMOTE demonstrated equivalent performance to using large-scale UK Biobank (UKB) data for addressing data imbalance.

Conclusions:

• TW-SMOTE is an effective method for resolving case-control data imbalance in human biobanks.
• TW-SMOTE offers a computationally efficient alternative to SAIGE for large genetic datasets.
• The developed TW-SMOTE approach supports precision medicine goals by improving genetic disease research with imbalanced data.