HawkFranklin Research

Derma AI

Internally: PelliScope Kiosk & Web Catalog

A state-of-the-art clinical decision support system. Validated across 10 dermatology conditions.

Narrative Direction

Guided by strategic insights from Oracle Health for maximum clinical relevance.

Developed For

Tailored specifically for the Emirates Health Services (EHS) ecosystem.

Global Showcase

Officially presented at Arab Health & WHX Health under the brand "Derma AI".

Diagnostic Excellence

Clinical Validation Benchmarks

10-Condition Panel

Eczema Psoriasis Tinea Herpes Zoster Allergic Dermatitis Irritant Dermatitis Insect Bite Urticaria Folliculitis Drug Rash

Multi-Class Diagnostic Accuracy

0.800

Macro ROC-AUC

0.849

Micro ROC-AUC

Outperforms general multimodal LLMs in zero-shot evaluations

Infectious Triage

Automatically screens and distinguishes between infectious and non-infectious conditions at the first point of contact. Protects clinical staff and prioritizes patient isolation.

Binary Screening Performance

Micro AUC
0.919
Macro AUC
0.814
Balance-Point Tuned Thresholds Robust across 10-seed multi-run

Technological Engine

Multi-Instance Learning Architecture

Visual Processing Pipeline

Attention-Based MIL

Real clinical encounters involve multiple photos. Derma AI moves past single-image classification by dynamically aggregating 1 to 3 patient views (Close-up, Angled, Contextual) into one unified, case-level prediction.

  • 6,144-D Derm Embeddings
    Built atop leading domain-tuned foundation models, fundamentally outperforming generic IMAGENET representations.
  • Interpretable Attention Weights
    Clearly identifies which exact image framing (e.g. Context instead of Close-Up) held the most diagnostic diagnostic weight.
Close-Up
Angled
Context

CPU-Optimized Inference

Zero reliance on high-cost GPUs. The system runs securely on standard on-prem computing hardware, yielding a ~6-7 second latency per multi-image case processing.

End-to-End Speed
6.34s Average Delay

Physical Kiosk Ready

EMBEDDED

PelliScope is packaged as an end-to-end interactive hardware Kiosk solution, crafted specifically for contactless, walk-up triage in hospital environments.

Also supports Web & Mobile Workflows

Commercial & Sovereign Economics

Pricing & Localization Strategy

Baseline Deployments

GLOBAL RATES

On-Premise Hospital Core

$15k / instance

Enterprise privacy; diagnostic models live within internal networks.

Tele-Medicine API License

$4k base + $0.50 / query

Hosted microservice API for high-volume virtual self-screening apps.

Hardware Resource Cost (6H Day) ~$2.50

Sovereign Data Adaptation

Indian Demographic Targeting

To comply with national AI governance laws and ensure accuracy against ethnic Fitzpatrick IV-VI skin tones, the architecture natively supports sovereign recalibration.

Local Model Training Fee
~$25k
One-Time Setup
  • Local sovereign dataset curation.
  • Hyperparameter & attention adjustments.
  • Balance-point thresholds recalibration.
EHS Letter of Acknowledgement

Physical documentation retained securely on-file.