Adapting DeepLabV3+ for biopsy cervical cancer lesion segmentation
-
By
-
Rose Nakasi
-
Cosmas Wamozo
-
Solomon Nsumba
-
Benjamin Rukundo
-
Tonny Okecha
-
Byron Mubiru
-
Chodrine Mutebi
-
May 11, 2026
-
Clinical Scorecard: Modifying DeepLabV3+ for the Segmentation of Biopsy Lesions in Cervical Cancer
At a Glance
| Category | Detail |
|---|
| Condition | Cervical Cancer |
| Key Mechanisms | DeepLabV3+ architecture with ResNet34 encoder for histopathological image segmentation. |
| Target Population | Patients with cervical cancer, particularly in resource-constrained settings. |
| Care Setting | Resource-limited healthcare environments, such as rural and remote areas. |
Key Highlights
- Achieved mean Intersection over Union (IoU) of 75.8% and Dice coefficient of 93.1%.
- Utilized smartphone-assisted microscopy for image acquisition.
- DeepLabV3+ outperformed U-Net baseline in segmentation tasks.
- Targeted 21 distinct histopathological feature classes.
- Demonstrated feasibility of digital pathology in low-resource settings.
Guideline-Based Recommendations
Diagnosis
- Utilize smartphone-assisted microscopy for standardized image acquisition.
Management
- Implement DeepLabV3+ for accurate segmentation of cervical cancer lesions.
Monitoring & Follow-up
- Evaluate segmentation performance using metrics like IoU and Dice coefficient.
Risks
- Potential limitations in generalization to broader clinical populations without multi-institutional validation.
Patient & Prescribing Data
Cervical cancer patients in low and middle-income countries.
Affordable digital pathology solutions can improve diagnostic access and treatment planning.
Clinical Best Practices
- Combine smartphone technology with deep learning for enhanced diagnostic capabilities.
- Focus on multi-scale feature extraction for complex histopathological patterns.
- Ensure rigorous validation of segmentation models on diverse datasets.
Related Resources & Content
