SMOTE-Based Automated PCOS Prediction Using Lightweight Deep Learning Models | JoVE Visualize

Area of Science:

Biomedical Engineering
Artificial Intelligence in Healthcare
Reproductive Health Research

Background:

Polycystic Ovarian Syndrome (PCOS) significantly impacts women of reproductive age, with high testosterone levels contributing to miscarriages and ovulation issues.
PCOS affects a substantial portion of the population, with recent studies indicating prevalence rates as high as 31.3% in Asian women.
Existing machine learning approaches for PCOS detection often rely on manual feature extraction, leading to performance limitations and hindering accurate diagnosis.

Purpose of the Study:

To develop and evaluate cutting-edge deep learning models for automated feature engineering in PCOS prediction.
To enhance the accuracy and performance of PCOS detection by leveraging advanced deep learning techniques.
To address the limitations of traditional machine learning methods in accurately identifying PCOS.

Main Methods:

Proposed three lightweight deep learning models: LSTM-based, CNN-based, and CNN-LSTM-based.
Utilized the Synthetic Minority Over-sampling Technique (SMOTE) for effective dataset balancing to ensure robust model performance.
Implemented automated feature extraction through deep learning architectures, eliminating the need for manual feature engineering.

Main Results:

The CNN-based model achieved the highest accuracy (96.59%) and ROC-AUC (96.6%), with a minimal number of parameters (297) and rapid training time (10.02 s).
Statistical significance was confirmed using the DeLong test, comparing the AUCs of all three models.
The SMOTE + CNN model outperformed other models in accuracy, precision, recall, AUC, parameter count, and training efficiency.

Conclusions:

The proposed deep learning models, particularly the CNN-based approach, demonstrate superior performance in PCOS detection compared to existing state-of-the-art methods.
The developed model shows potential for early PCOS identification, which can contribute to reducing pregnancy complications such as miscarriages.
This research highlights the efficacy of deep learning with automated feature engineering for improving PCOS diagnosis and management.