MLNet: Metaheuristics-Based Lightweight Deep Learning Network for Cervical Cancer Diagnosis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a leading cause of cancer deaths in women, necessitating early diagnosis.
• Current deep learning models for cervical cancer diagnosis face challenges like overfitting and parameter tuning.
• Automated diagnosis systems are crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To propose a novel metaheuristics-based lightweight deep learning network (MLNet) for automated cervical cancer diagnosis.
• To address limitations of existing deep learning models, including overfitting and gradient vanishing.
• To optimize Convolutional Neural Network (CNN) architecture and hyper-parameters using advanced metaheuristic algorithms.

Main Methods:

• Defined CNN hyper-parameter tuning as a multi-objective problem.
• Employed Particle Swarm Optimization (PSO) for optimal CNN architecture design.
• Utilized Dynamically Hybrid Niching Differential Evolution (DHDE) for CNN layer hyper-parameter optimization.
• Trained and validated MLNet on three benchmark cervical cancer datasets (Herlev, SIPaKMeD, Mendeley LBC).

Main Results:

• MLNet demonstrated superior performance across all tested datasets compared to existing models.
• Significant improvements were observed in accuracy, f-measure, sensitivity, specificity, and precision.
• For instance, on the Herlev dataset, MLNet achieved accuracy improvements of 1.6254% and f-measure improvements of 1.5178%.

Conclusions:

• The proposed MLNet effectively overcomes common deep learning challenges in cervical cancer diagnosis.
• MLNet offers a robust and accurate solution for automated cervical cancer detection.
• The metaheuristics-driven approach enhances the reliability and performance of deep learning models in medical diagnostics.