Sky is the limit

It is all about integraition of different technologies

Image recognition

Feature labeling.

Neural Networks

Pattern recognition

Hardware

Macro photography

Breakthroughs often emerge from the integration of

And this is our case.

Before exploring the technologies, let’s have a look how skin cancer pattern recognition is done by humans.

The Cells

Melanocytes

Why are they so different from other skin cells?

Special kind of skin cells that produce melanin – dark pigment that protects our skin from the sun radiation.

andrey-zvyagintsev-byfNhh81CWU-unsplash-bw

Before

After

What is Skin Cancer

DNA damaged cells

At the same time melanocytes are source of the most dangerous form of skin cancer – Melanoma.

Skin cancer cells are DNA damaged cells that, unlike other form of cancer, are visible under certain conditions.

Taking picture

Dermatoscopy

Usually, doctors take a tissue picture with Dermatoscope, special equipment with magnifying glass that allows to make macro photo in natural and polarized light with different scales and filters.

Picture analysis

Risk Assessment

Then image is labeled based on one of the pattern recognition methods. Some of the most popular are: ABDCE, Seven Point Check-list, Menzies methos. All of them are slightly different, but all of them use the same visual color, structure and pattern recognition techniques.

HOW TECHNOLOGIES CAN HELP?

The Core Technology

Convolutional Neural Network

At the heart of our skin cancer detection system lies Convolutional Neural Networks (CNNs), a class of deep learning models tailored for image recognition tasks. CNNs excel at analyzing visual data, making them particularly effective for medical image analysis, such as identifying skin abnormalities. By training our CNN on thousands of dermoscopic images of both benign and malignant lesions, our technology learns to differentiate between healthy skin and potential cases of skin cancer, including melanoma.

Evolution

Self learning

Our CNN model processes images by breaking them down into layers, analyzing features like color, texture, and patterns that are indicative of skin cancer. The network identifies subtle differences in visual markers, which may be hard for the human eye to detect, and applies this knowledge to classify new, unseen images.

Over time, with continuous training and improvements, our system becomes even more adept at recognizing complex and rare skin cancer cases, ensuring higher accuracy and reliability.

WHAT ABOUT TRAINING DATA?

Training

Data sets

To enhance the accuracy and robustness of our skin cancer detection system, we train our CNNs using a variety of data sources, a mix of public datasets, clinical data, and crowdsourced images.

Public datasets

Good to start

Publicly available datasets contain thousands of annotated images of skin lesions. These datasets provide a strong foundation for training the model to recognize different types of skin abnormalities, including benign, malignant, and atypical lesions. Public datasets are essential for establishing a broad baseline of knowledge, enabling the CNN to learn from a wide array of cases.

One of the largest skin cancer image datasets is collected by The International Skin Imaging Collaboration (ISIC) and it keeps 482,781 images.

About ISIC

The International Skin Imaging Collaboration (ISIC) is a global initiative aimed at improving the diagnosis of skin cancer through the development of standardized imaging and machine learning technologies.

ISIC focuses on creating and sharing a vast repository of dermoscopic images, which are used by researchers and medical professionals to advance skin cancer detection.

By fostering collaboration among dermatologists, researchers, and AI experts, ISIC helps to promote the development of accurate, AI-driven diagnostic tools. Their efforts are critical in enhancing early detection and treatment of skin cancers, including melanoma, worldwide.

Clinical data

Good for labelling and calibration

We have partnered with two dermatology clinics and engaged 12 dermoscopy-certified dermatologists to help us calibrate and refine our model for skin cancer detection. The clinical data is use both for training the model and fine-tuning its output for greater accuracy.

Pilot testing

Our model is tested in the clinics to verify the results, ensuring the system performs reliably in clinical practice. This collaboration helps us continuously improve the model’s diagnostic precision and usability.

Crowdsourcing

For precision

We are developing a mobile application that enables users to perform an initial risk assessment for skin cancer by analyzing images of their skin lesions. With the user’s consent, the images can be reviewed and labeled by dermoscopy-certified dermatologists, ensuring expert input.

More training data

These labeled images not only provide users with a more accurate evaluation but also contribute valuable data to further improve our AI model. By incorporating real-world feedback, we continually enhance the app’s diagnostic accuracy and effectiveness in detecting skin cancer.

Data quality

Calibration and mapping

Since our datasets come from various sources — public datasets, clinical data, and crowdsourced images — they differ in structure, resolution, and classification types. To ensure these diverse datasets are suitable for training our AI model, we perform extensive pre-processing. This involves standardizing image formats, adjusting resolutions, and harmonizing classification labels to create a consistent dataset. By carefully calibrating these qualified datasets, we ensure the model can learn effectively and produce accurate results across all input types.

HOW TO GET DERMATOSCOPY IMAGE?

Device

Dermatoscope

A dermatoscope is a medical device used for examining skin lesions with enhanced magnification and illumination. It allows dermatologists to observe the skin in greater detail, revealing structures beneath the surface that are not visible to the naked eye.

By using polarized or non-polarized light, dermatoscopy helps in diagnosing skin conditions like melanoma, basal cell carcinoma, and other skin abnormalities.

Can you do it with you Iphone?

Macrophotography

In 2021, Apple released the iPhone 13 Pro with advanced lens technology that enabled macro photography, setting a new standard for flagship models across other smartphone brands. Modern smartphone cameras now feature lenses and sensors capable of capturing high-quality close-up images with exceptional detail.

With the right lighting conditions, these cameras can produce skin images of sufficient quality to support dermatoscopic assessments. Combined with specialized apps and AI-driven tools, this technology allows for convenient documentation and analysis of skin conditions, offering an accessible, portable alternative to traditional dermatoscopes for both users and healthcare professionals.

WE ARE EXCITED ABOUT THE OPPORTUNITY TO CREATE CONVENIENT TOOL FOR SKIN CANCER ASSESSMENT