As many as 90,000 new melanoma cases are diagnosed each year. Out of these, 9,000 are expected to die of malignant melanoma. Furthermore, the rates of melanoma have been rising for the last thirty years.

Melanoma is a dangerous form of skin cancer in which cancerous growths resembling moles develop on skin, often spreading to other parts of the body. This type of cancer is very difficult to detect as patients do not consistently track changes on skin and therefore, are unable to monitor lesions over time and note differences in morphology. There is a large need in the community for a scalable, easy-to-use platform that allows patients to take their health into their own hands and provide preliminary diagnoses for melanoma. Additionally, when a patient is in between annual checkups or is unable to visit the doctor, such a platform would allow patients to increase their own role in monitoring their health and in turn, report back to physicians.

In this generation, smartphones are becoming more and more prevalent worldwide, allowing access to even the most remote countries. This relatively cheap and largely accessible technology is the ideal platform to use to allow customers around the world to easily prevent and aid in the early diagnosis of melanoma.

What it does

We created a webapp with a supplementary mobile app that captures a picture of the lesion of interest on the body. This lesion image is tested using a custom-vision API machine learning algorithm that calculates the risk of melanoma based on four factors: lesion asymmetry, lesion border irregularity, lesion color, and lesion surface.

The app returns a summary of the calculated risk of melanoma for each factor with a recommended line of action. The calculated risk percentage is divided such that 0-40% is considered low risk for melanoma, 41-60% is intermediate risk, and 61-100% is significant risk for melanoma based on a specific factor. A final calculated average risk of the percentages of all factors is also presented with recommended next steps. Low risk results are advised to monitor lesions over time, intermediate risks are advised to monitor lesions closely and visit a physician soon, and significant risk results are advised to make an appointment with a physician as soon as possible.

This machine learning algorithm for identification of melanoma has a 92.4% accuracy. This information is meant to enable the patient to monitor their lesions and use algorithm results to supplement their decision to visit a physician.

How we built it

Web App/Website using Java, Javascript, HTML, and CSS - Melanoma identification algorithm development through Microsoft's Custom Vision API.

Challenges we ran into

We had difficulty transferring ML results from the API to our webapp.

Accomplishments that we're proud of

We achieved 92.4% accuracy on the identification of melanoma.

What we learned

Because smartphones are accessible to most in society, they have become the ideal technology to pair with machine learning to create a smart system for real-time monitoring of lesions. This telemedicine platform allows patients to act as soon as abnormalities become apparent.

What's next for Melanotix

We hope to further tune our machine learning algorithm by implementing weighted averages based on the importance of each lesion factor on risk of melanoma. In addition, we hope to foster communication between physicians and patients through a platform in the webapp to facilitate direct conversation based on algorithm results. Finally, we plan to partner with various telehealth platforms to allow for a more holistic user experience and to integrate lesion progression over time into the platform.

+ 1 more
Share this project: