Booth Id:
SOFT041T
Category:
Systems Software
Year:
2021
Finalist Names:
Bodepudi, Asritha (School: Lexington High School)
Meel, Vidushi (School: Lexington High School)
Abstract:
Malignant melanoma, a common skin cancer caused by integumentary cells called melanocytes, has a median survival rate of 5.3 months and less than 15% five-year survival rate for patients post metastasis [18]. Metastasis describes the state of cancer where secondary malignant growths develop in other organs away from the primary site. Identifying melanoma before metastasis will increase the probability of successful removal via excision or other minimally invasive techniques. This paper presents Melatect, a machine learning model created to allow users to identify potential malignant melanoma. A recursive computer image analysis algorithm was used to create a machine learning model which is capable of detecting likely melanoma. The comparison is performed using 20,000 raw images of benign and malignant lesions from the International Skin Imaging Collaboration (ISIC) archive that were augmented to 60,000 images. Tests of the algorithm using subsets of the ISIC images suggest it accurately classifies lesions as malignant or benign over 98% of the time with no apparent bias or overfitting. The Melatect iOS app was later created and published, in which the machine learning model was embedded. Users take pictures of skin lesions (moles) using the app, which are then compared to a database of malignant and benign lesions in order to identify it. Melatect provides a convenient way to get free advice on potentially malignant lesions and track these lesions over time. It is currently in clinical trials with dermatologists for more data on its accuracy.