Team: 1017
School: Los Alamos Mid
Area of Science: Computational Medicine
Interim: Melanoma cancer is a mole that is irregular to standard patterns like cell shape and structure along with color and many other things. It is visually detected, some of the visual signs of Melanoma are “A-Asymmetryâ€, “B-Border irregularityâ€, “C-Color changesâ€, “D-Diameterâ€, and “E-Evolving.†A-B-C-D-E isn’t a definitive way to see if it is or isn’t cancer but rather if one needs a biopsy to further find out. The melanoma tumors are usually brown due to the making of melanin. They can be other colors like pink and white. [ref 1] Just remember A-B-C-D-E! Melanoma is a curable disease if found early enough. If left too long the cancer will grow deeper and spread throughout your body. After this, the cancer continues to grow uncontrollably. About 96,480 new melanoma cases will be found in 2019. [ref 2] The average age of people diagnosed with melanoma is 65 years of age. [ref 2] About 7,230 people in that same year will die of it. [ref 2].
Because Melanoma is visually detected, artificial intelligence (AI), or “machine learningâ€, is a way to make a computer potentially detect it. AI can learn like a human brain and be able to recognize complex and subtle patterns. This will make it easier, and cheaper to look at moles and find if they are or aren’t regular. If irregular, it would require a biopsy to confirm it’s cancer because A-B-C-D-E doesn't definitely mean cancer. To do this we will train the AI to classify images based on training sets of healthy and cancerous moles.
AI uses a neural net. The inputs to the neural net are multiplied by the weight with each input going to each layer. This keeps happening until the output. At each node in a layer and the output, the sigmoid function is used to make the number between 0 and 1. We will use the Keras library for the implementation of the neural net. We will follow an existing example in the Keras documentation for image classification.
References:
1. Mayo clinic
2. American cancer society
3. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions
Mentors:
Ed McKigney
Nathaniel Morgan
Team Members:
Sponsoring Teacher: Aik-Siong Koh