Explainable clinical diagnosis through unexploited yet optimized fine-tuned ConvNeXt Models for accurate monkeypox...

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Healthcare
• Deep Learning Applications

Background:

• Deep learning (DL) excels in healthcare image processing, but monkeypox detection is challenging due to symptom overlap with chickenpox and measles.
• The global spread of monkeypox necessitates rapid and accurate diagnostic tools.
• Existing DL models for skin conditions often require extensive resources, limiting real-time use.

Purpose of the Study:

• To develop an efficient and accurate deep learning framework for monkeypox detection using visual data.
• To leverage transfer learning (TL) and ConvNeXt architectures to overcome computational limitations of traditional DL models.
• To evaluate the performance and interpretability of the proposed model for clinical decision-making.

Main Methods:

• Utilized transfer learning (TL) to fine-tune pre-trained ConvNeXt networks (ConvNeXtSmall, ConvNeXtBase) for monkeypox classification.
• Implemented various pre-processing and data augmentation techniques, optimizing for performance and computing time.
• Employed Adafactor optimization and assessed models using standard train-test splits and k-fold cross-validation.

Main Results:

• Achieved 99.9% accuracy on the binary-class MSLD dataset and 94% accuracy on the multi-class MSLD v2.0 dataset.
• Demonstrated the practicality and efficiency of the TL-based ConvNeXt framework.
• Incorporated explainable AI methods for enhanced model interpretability.

Conclusions:

• The proposed TL-based ConvNeXt framework provides a highly accurate and computationally efficient solution for monkeypox detection.
• The model's explainability aids healthcare professionals in understanding diagnostic decisions.
• This approach offers a viable alternative to existing methods, addressing real-time applicability concerns in public health.