Skin Spot AI - The Science

Skin lesion risk assessment using your smartphone.

Join waitlist

The science behind Helfie's Skin Spot AI

Helfie's Skin Spot AI turns a compatible smartphone into an intelligent screening tool that identifies and analyses skin lesions for early risk assessment.

By combining computer vision, deep learning, and real-world dermatology data, Skin Spot AI provides fast, non-invasive screening to support early detection and preventive care.

Using a smartphone camera, the system analyses key visual characteristics of skin lesions, including asymmetry, border irregularity, colour complexity, and structural patterns, to assess whether a lesion presents a low, medium, or high risk.

Why is skin screening important?

Skin cancer is one of the most common cancers worldwide, affecting more than 1.5 million people each year⁷. Many skin lesions develop without noticeable symptoms, making early recognition particularly important⁸.

Access to dermatologists remains limited by long wait times, high costs, and specialist shortages, particularly in rural and underserved communities⁹. These barriers can delay diagnosis and treatment, increasing the risk of avoidable disease progression¹⁰.

Helfie's Skin Spot AI helps address this gap by providing immediate, low-cost screening using a smartphone. By identifying potentially concerning lesions, it can support timely clinical follow-up and help people monitor changes in their skin from home. This is particularly transformative for rural and underserved populations with limited access to dermatology services.

Sources
7 - IARC / WHO: Skin cancer

8 - The Skin Cancer Foundation: Skin cancer facts & statistics

9 - Dermatology Times: Closing the gap: Why rural dermatology access must be a priority

10 - JMIR Dermatology (2025): Understanding the dermatology accessibility crisis

How does a smartphone camera assess skin lesions?

Skin Spot AI uses a two-stage deep learning process to detect skin lesions and assess their potential risk.

1. Lesion detection

High-resolution smartphone images are analysed using a convolutional neural network (CNN) trained to locate skin lesions. The model identifies visual features such as asymmetry, border shape, colour variation, and texture to isolate the lesion for assessment.

2. Risk classification

The isolated lesion is processed by a DINOv3-based vision transformer, fine-tuned on 2,298 annotated dermatology images. The model classifies lesions into three risk categories:

  • Low risk: Benign lesions (e.g. nevi, seborrhoeic keratosis).
  • Medium risk: Actinic keratosis.
  • High risk: Malignant lesions (melanoma, BCC, SCC).

This architecture uses attention mechanisms to identify clinically meaningful visual features and deliver a risk estimate within seconds.

Is Helfie's skin screening independently validated?

Skin Spot AI has demonstrated strong performance across a real-world dermatology dataset:

Reported screening performance:

  • 83.7%

    Overall accuracy

  • 94.8%

    AUC

  • 85.3%

    High-risk accuracy

  • 94.5%

    Low-risk accuracy

Sensitivity, specificity, F1 scores, and a Cohen’s kappa of 0.741 indicate strong agreement with established clinical screening benchmarks. Performance remained stable across varied skin tones and device conditions, supporting reliable everyday use.

Join waitlist Learn more

Ready to get Helfie?

Take control of your health and join the waitlist.

Helfie Licensing Pty Ltd will use your email for the purpose of providing you with further information about Helfie and to give you notice when early access is available. You can find out more about how we manage your data in our Privacy policy.