Detection of malignant melanomas using neural networks

Examensarbete på kandidatnivå

Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.12380/304921
Download file(s):
File Description SizeFormat 
22-10 Länsberg Manfredsson.pdf6.8 MBAdobe PDFView/Open
Bibliographical item details
FieldValue
Type: Examensarbete på kandidatnivå
Title: Detection of malignant melanomas using neural networks
Authors: Länsberg, Simon
Manfredsson, Anna
Abstract: Approximately 60 000 people in Sweden are diagnosed with skin cancer each year, and around 500 of these patients die from their disease. There has been an increasing number of skin cancer cases in Sweden every year. This coupled with the highly stressed health care industry may result in a significant increase in mortality rates. Therefore, in an effort to detect possible malignant lesions on the skin, an image classification model was developed. The model in question was a convolutional neural network, a type of deep learning that specialises in classifying image data. In order to construct the dataset we used images found in the ISIC archives and divided them into two classes, malignant and benign. Several attempts were made before the best model was developed with a combination of transfer learning and the loss function ADAM. The model demonstrated an average performance of 73%. Using the Flutter framework it was possible to build an accompanying application with which the model could be presented to the general public. Ultimately, the app provided its users with the ability to take a picture of their lesion and then receive an indication based on the recommendation provided by the model. The connection between the application and the model was made possible through a Firebase database and a Python script that housed the model.
Keywords: AI;Melanoma;CNN;Transfer Learning;Python;Flutter;Firebase
Issue Date: 2022
Publisher: Chalmers tekniska högskola / Institutionen för data och informationsteknik
URI: https://hdl.handle.net/20.500.12380/304921
Collection:Examensarbeten för kandidatexamen // Bachelor Theses



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.