The Problem

Skin cancer affects millions worldwide, yet access to dermatological screening remains limited. Long wait times, geographic barriers, and the high cost of specialist visits mean many suspicious lesions go unexamined until it is too late.

Melanoma caught early has a 99% survival rate; caught late, that drops to 25%. The gap between these outcomes is often simply access to timely screening.

Our Solution

NOMA AI is a portable, AI-powered skin cancer screening system designed to democratize dermatological assessment. Built on a Raspberry Pi 4 with a custom enclosure, it integrates:

• AI Visual Analysis
  A MobileNetV3 deep learning model trained on 12,900 images across 24 skin conditions (4 malignant, 20 benign)

• Clinical Risk Assessment
  The ABCDE framework (Asymmetry, Border, Color, Diameter, Evolution) used by dermatologists worldwide

• Explainable AI
  Grad-CAM heatmaps that highlight exactly which image regions influenced the model’s decision

• Immediate Physical Feedback
  A tri-color LED system (Red / Yellow / Green) for instant risk communication

Technical Architecture

Hardware

• Raspberry Pi 4 (4GB RAM) — Real-time inference and system control

• Arducam IMX519 (16MP) — High-resolution autofocus imaging

• 5" Waveshare Touchscreen — Interactive GUI with swipe navigation

• Custom LED Array — Red / Yellow / Green indicators

• Portable Power Bank — Fully mobile, clinic-ready design

Software Pipeline

1. Image Capture — High-resolution lesion photography with real-time preview

2. AI Inference — TFLite-optimized MobileNetV3 running entirely on-device

3. Feature Extraction — Automated analysis of asymmetry, border irregularity, color uniformity, and diameter

4. Clinical Assessment — Step-by-step ABCDE wizard with patient history

5. Risk Fusion — Weighted combination of AI confidence, ABCDE score, and patient risk factors

6. Interpretability — Grad-CAM heatmap generation to visualize AI attention regions

Model Training Details

Dataset
12,900 images across 24 classes, compiled from public dermatology repositories.

Architecture

MobileNetV3 (ImageNet pre-trained) with a custom classification head:

• Global Average Pooling

• Dropout (0.3)

• Dense (512, ReLU) + BatchNorm

• Dropout (0.4)

• Dense (256, ReLU)

• Dropout (0.3)

• Softmax (24 classes)

Training Strategy

• Stage 1: Classifier training (30 epochs, LR = 0.001)

• Stage 2: Fine-tuning last 40 layers (40 epochs, LR = 0.0001)

• Stage 3: Full model training (30 epochs, LR = 0.00001, if needed)

Class Balancing

Automatic class weighting using scikit-learn to address dataset imbalance.

Optimization

Converted to TensorFlow Lite using tf.lite.Optimize.DEFAULT for efficient edge deployment.

Risk Calculation

NOMA AI combines multiple factors to generate a comprehensive risk score:

ComponentWeightDescription

AI Confidence40%Model prediction with confidence score

ABCDE Score40%Clinical features (Asymmetry, Border, Color, Diameter, Evolution)

Patient Risk20%Age, skin type, family history, sunburn history

LED Alert System

• 🔴 RED (≥70%) — High risk → Urgent dermatology referral

• 🟡 YELLOW (40–69%) — Moderate risk → Schedule follow-up

• 🟢 GREEN (<40%) — Low risk → Continue self-monitoring

Recent Updates (March 2026)

The latest version of NOMA AI includes major enhancements based on testing feedback:

• Top-3 Alternative Diagnoses — Improves interpretability by showing other likely conditions

• Uncertainty Estimation — Flags predictions with >80% uncertainty, prompting retakes

• Clinical Feature Extraction — Automated scoring of asymmetry, border, and color

• Educational Tip Banner — Random dermatology insights to promote skin health awareness

• Health Passport — Stores assessments in JSON format for longitudinal tracking

• Expanded Clinical Context — New inputs: itchiness, sudden onset, recurrence

These updates elevate NOMA AI from a simple classifier to a clinical decision support system.

Project Goals

NOMA AI was designed to:

1. Democratize access to preliminary skin cancer screening

2. Bridge the gap between AI automation and clinical reasoning

3. Educate users about skin health and warning signs

4. Provide explainable results that build trust

5. Enable longitudinal health tracking through a “health passport”

Limitations

NOMA AI is a screening tool — not a diagnostic device. It does not replace:

• Professional dermatological examinations

• Biopsy and histopathology

• In-person medical consultation

Always consult a healthcare professional for medical concerns.

The Team

NOMA AI was developed as a CYSF 2026 project by a student passionate about accessible healthcare technology.

The system integrates machine learning, embedded systems, and clinical reasoning into a portable, affordable screening solution.