Advancing Prostate Cancer Diagnostics: A ConvNeXt Approach to Multi-Class Classification in Underrepresented...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Prostate cancer poses a significant global health burden, with diagnostic disparities in underrepresented regions like sub-Saharan Africa.
• Limited availability of diverse datasets hinders equitable advancements in AI-driven cancer diagnostics.

Purpose of the Study:

• To develop and validate a novel ConvNeXt deep learning model for multi-class classification of prostate histopathological images.
• To address the lack of regional representation in diagnostic datasets by utilizing data from Nigeria and the ProstateX dataset.

Main Methods:

• A ConvNeXt architecture was employed for classifying prostate histopathology images into normal, benign, and malignant categories.
• The model incorporated advanced data augmentation, Grad-CAM for interpretability, and ablation studies for optimization.
• Performance was evaluated against traditional CNNs and transformer models, and validated on the ProstateX dataset.

Main Results:

• The ConvNeXt model achieved 98% accuracy on the Nigerian dataset, outperforming ResNet50, EfficientNet, DenseNet, ViT, CaiT, Swin Transformer, and RegNet.
• Validation on the ProstateX dataset yielded 87.2% accuracy, 85.7% recall, 86.4% F1 score, and 0.92 AUC.
• Grad-CAM visualizations provided clinical explainability, and ablation studies confirmed the model's robustness.

Conclusions:

• ConvNeXt demonstrates superior performance in prostate cancer histopathology classification, offering a robust and interpretable solution.
• The study highlights the potential for AI to improve cancer diagnostics in low-resource settings and underrepresented populations.
• This work advances equitable healthcare by promoting inclusivity in global cancer diagnostic research.