Modeling Deep Learning Neural Net Based Image Recognition to Classify Melanoma Better than a Trained Professional
CSEF · 2016 Microbiology (General)
Overview
Objectives/Goals The purpose of this experiment was to determine if the latest artificial intelligence (deep learning neural net) image recognition algorithms had reached a level of sophistication to create a melanoma classification model (deep learning classifier) that could distinguish melanoma from other forms of skin cancer (non-melanoma) more accurately than a trained professional#s classification mental model (human classifier). Methods/Materials Materials included a Windows laptop, 800 images of melanoma and non-melanoma scans collected from Lloyd-Derm, and Metamind's generic deep learning based image recognition algorithm. 640 images were used to train the computer algorithm and the remaining 160 images were used to determine the accuracy of the training. A subset of these 160 images was to a human classifier (an oncologist, a trained professional) for her to classify. Results After the deep learning classifier was trained to identify melanoma, its accuracy was compared to that of a human classifier with a controlled test sample. It was found that deep learning classifier had an accuracy of 85% and the human classifier had an accuracy of 68%. Conclusions/Discussion The deep learning classifier achieved a higher accuracy than the human classifier by a significant margin. Every year, skin cancer affects 5.4 million people in the US and costs $8.1 billion in treatment as the diagnosis of the disease costs up to $10000. The survival rate reduces from 94% to 15% when detected later. Detecting melanoma using the deep learning classifier is more accurate and quicker, thus positively impacting both the survival rate and the overall cost of diagnosis.
Summary statement
Using a generic deep learning based image recognition algorithm, I created a classifier to distinguish melanoma from other forms of skin cancer with a higher accuracy than a trained oncologist.
Help received
Dr. Swarajya Lakshmi Vemuri from Kaiser Permanente classified the medical images with her experience as an oncologist. Metamind's generic deep learning algorithm and images from Lloyd-Derm, DermNet NZ, and DermIS were used in this project.
Competition history
- CSEF 2016
Resources
Related projects
ISEF · 2016
Novel Machine Learning Algorithms for Early Diagnosis of Melanoma
ISEF · 2017
Utilizing Machine Learning Techniques to Identify Cancerous Skin Lesions
ISEF · 2024
Detect Early Melanoma Cancer Using Machine Learning
ISEF · 2018
Detection of Melanoma via Deep Learning
ISEF · 2017
A Novel Machine Learning Approach Using Convolutional Neural Networks to Identify Melanoma
CSEF · 2014
Accurate Detection of Skin Cancer Using Multi Stage Neural Networks
CSEF · 2018
skinCAM: An Effective Solution towards Skin Disease Diagnosis via Novel Deep Learning Algorithm
ISEF · 2022
A Multi-Output Convolutional Neural Network Model for Melanoma Detection and Prevention
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects