Novelty Classification Model Use in Reinforcement Learning for Cervical Cancer

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a major global health concern.
• Early detection is crucial for effective patient outcomes.
• Class imbalance in datasets often hinders diagnostic model accuracy.

Purpose of the Study:

• To enhance cervical cancer diagnosis accuracy.
• To develop a novel hybrid deep learning model.
• To address and mitigate class imbalance in medical image classification.

Main Methods:

• A hybrid model, RL-CancerNet, integrating EfficientNetV2 and Vision Transformers (ViTs) within a Reinforcement Learning (RL) framework.
• EfficientNetV2 for local feature extraction; ViTs for global dependency recognition.
• An RL agent dynamically adjusts focus to minority classes; a Supporter Module with Conv3D, BiLSTM, and attention enhances contextual learning.

Main Results:

• RL-CancerNet achieved 99.7% accuracy on benchmark datasets (Herlev and SipaKMeD).
• Performance surpasses existing state-of-the-art models.
• Demonstrated effectiveness in identifying subtle features in complex backgrounds.

Conclusions:

• The integration of CNNs, ViTs, and RL in RL-CancerNet significantly boosts diagnostic accuracy for cervical cancer screenings.
• The model represents an advancement in automated medical screening.
• Offers a scalable framework adaptable to other medical imaging tasks.