A Comprehensive Diagnostic Tool for Skin Cancer Using a Multifaceted Computer Vision Approach
This paper introduces a system that has the capability to effectively diagnose a skin lesion through machine learning computer vision techniques that enable realtime assessment. This enables individuals to efficiently diagnose themselves and get the necessary treatment quickly. In addition, this system is also available for clinics, permitting easy and inexpensive diagnosis, as opposed to traditional diagnosis methods such as Magnetic Resonance Imaging devices (MRI), or Computed Tomography scans (CT). We design an approach that combines Object Detection with an ensemble of Convolutional Neural Networks (CNN) and encapsulates it in a button based user-navigable Graphical User Interface (GUI), leveraging the video input from a user’s webcam (either built-in or external) to locate and identify the lesion instantaneously. Additionally, the tool provides critical information about the diagnosis, as well as links to medical resources and treatment options. We achieve a top-l accuracy of 87% $\pm 2.0\%$, pointing to our tool as a viable and highly accurate self-diagnosis strategy for individuals globally. Our system can potentially reach those in areas where access to diagnostic tools and healthcare professionals is seriously lacking.
Paper
Full text
A Comprehensive Diagnostic Tool for Skin Cancer Using a Multifaceted Computer Vision Approach
Semantic Scholar · Medicine · 2020
Abstract
This paper introduces a system that has the capability to effectively diagnose a skin lesion through machine learning computer vision techniques that enable realtime assessment. This enables individuals to efficiently diagnose themselves and get the necessary treatment quickly. In addition, this system is also available for clinics, permitting easy and inexpensive diagnosis, as opposed to traditional diagnosis methods such as Magnetic Resonance Imaging devices (MRI), or Computed Tomography scans (CT). We design an approach that combines Object Detection with an ensemble of Convolutional Neural Networks (CNN) and encapsulates it in a button based user-navigable Graphical User Interface (GUI), leveraging the video input from a user’s webcam (either built-in or external) to locate and identify the lesion instantaneously. Additionally, the tool provides critical information about the diagnosis, as well as links to medical resources and treatment options. We achieve a top-l accuracy of 87% $\pm 2.0%$, pointing to our tool as a viable and highly accurate self-diagnosis strategy for individuals globally. Our system can potentially reach those in areas where access to diagnostic tools and healthcare professionals is seriously lacking.