To develop an automated histopathological image segmentation solution for cervical cancer lesions using affordable smartphone-assisted microscopy and deep learning, addressing the diagnostic gap in resource-limited settings.
Key Findings:
Achieved a mean IoU of 75.8% and a Dice coefficient of 93.1% on validation data, indicating strong segmentation performance.
Per-class IoU ranged from 74.13% to 75.41% across 21 feature classes, demonstrating consistent segmentation performance.
DeepLabV3+ outperformed U-Net baseline (mIoU: 56.84%, Dice: 68.53%), highlighting the effectiveness of the model architecture.
Interpretation:
The results indicate that smartphone-assisted microscopy combined with DeepLabV3+ can effectively segment cervical cancer lesions, demonstrating potential for improving diagnostic access and patient outcomes in resource-limited settings.
Limitations:
Validation results are based on a single institutional dataset, which may limit the generalizability of the findings.
Further independent multi-institutional evaluations are needed to assess the model's performance across diverse clinical populations and imaging conditions.
Conclusion:
This study presents a feasible approach to enhance cervical cancer diagnostics in low-resource environments, leveraging affordable technology and advanced deep learning.
