Momahmoses/ng-counterfeit-drug-detection
Domain:
healthcareRecord type:
software
Creator:
Mom
Host:
||Multi-modal counterfeit drug detection for Nigeria using NIR spectroscopy, pill vision, PaddleOCR packaging verification and NAFDAC registry lookup for offline Android deployment.
# Counterfeit Drug Detection, AI for Nigerian Medicine Markets
> Multi-modal ML system combining NIR spectroscopy, pill image analysis, and packaging OCR to identify counterfeit and substandard drugs in Nigerian markets, deployable as an offline Android app with a clip-on NIR spectrometer. Targeting the 42% of Nigerian drugs estimated to be falsified or substandard (WHO).
---
## The Problem
The WHO estimates **42% of antimalarials and antibiotics** in Nigerian markets are substandard or falsified. A community health worker has no way to verify drug authenticity in the field without a lab. Counterfeit drugs kill directly through treatment failure and contribute to antimicrobial resistance. The entire Nigerian supply chain, from port to patent medicine vendor, lacks a real-time authentication layer.
---
## Solution: Three-Modality Detection Pipeline
| Modality | Technology | What It Detects |
|---|---|---|
| **NIR Spectroscopy** | PLS-DA + CNN-1D | Active pharmaceutical ingredient (API) presence and concentration |
| **Pill Vision** | EfficientNet-B0 | Shape irregularity, colour deviation, imprint mismatch, surface defects |
| **Packaging OCR** | PaddleOCR + NAFDAC API | Fake NAFDAC numbers, misspellings, format violations, expired batch logic |
**Fusion layer**: weighted ensemble → `AUTHENTIC / SUSPICIOUS / LIKELY COUNTERFEIT`
---
## Priority Drug Coverage
- Artemether-Lumefantrine (ACT), most counterfeited antimalarial
- Amoxicillin, most counterfeited antibiotic
- Oxytocin, supply chain integrity critical for maternal survival
- ARV medications, HIV treatment efficacy
- Metformin / Glibenclamide, diabetes management
---
## System Architecture
```
[Field worker scans drug]
↓
[MODALITY 1: Clip-on NIR spectrometer (SCiO/Tellspec)]
→ SNV preprocessing → PLS-DA + CNN-1D
→ API present? Correct concentration?
[MODALITY 2: Camera photo of pill]
→ EfficientNet-B0 multi-label visual classifier
→ Shape / colour / imprint / surface texture score
[MODALITY …