BACK_TO_PORTFOLIO
// SYS.ID: 002
AI / ML

DIABETIC FOOT
ULCER CLASSIFIER_

CNN-based image classification system for automated detection and severity grading of diabetic foot ulcers, utilizing advanced preprocessing pipelines to improve model robustness and generalization.

VIEW_ON_GITHUB
CNN
Architecture
5
Grade Levels
92%
Precision
Python
Core
// OVERVIEW

Diabetic foot ulcers (DFUs) affect approximately 15% of diabetes patients and are a leading cause of non-traumatic lower limb amputations. Early, accurate classification of ulcer severity is critical to enabling timely treatment.

This system uses a custom CNN pipeline to classify wound images into severity grades (following the Wagner scale). Emphasis was placed on robust preprocessing — handling variable lighting conditions, wound occlusion, and wound orientation — to ensure the model generalizes to real-world clinical photography.

// TECH_STACK
Model Pipeline
  • → Python (core language)
  • → Custom CNN architecture
  • → OpenCV image preprocessing
  • → NumPy data handling
  • → Scikit-learn metrics / evaluation
Preprocessing
  • → CLAHE contrast enhancement
  • → Color space normalization (LAB)
  • → Wound region segmentation
  • → Rotation / flip augmentation
  • → K-fold cross-validation
// WAGNER_GRADE_SCALE
G0Intact skin with high-risk features (callus, deformity)
G1Superficial ulcer — no subcutaneous tissue involvement
G2Deep ulcer — tendon, capsule, or bone exposed
G3Deep ulcer with abscess or osteomyelitis
G4+Partial or full gangrene of forefoot / foot
// CHALLENGES & SOLUTIONS
01.
Inconsistent Image Quality — Clinical wound photos have highly variable lighting, angles, and occlusion. Applied CLAHE and color normalization to standardize inputs at inference time.
02.
Small Dataset — DFU datasets are limited due to patient privacy. Combated with aggressive augmentation and transfer-learning-inspired weight initialization.
03.
Clinical Reliability — Medical models must fail safely. Implemented confidence thresholding: predictions below 70% confidence are flagged for human review instead of being returned directly.
Python CNN OpenCV NumPy CLAHE Medical Imaging Wagner Scale Classification