Machine learning decomposition models for partial ordering problems: An application to melanoma severity classification

Research areas:

• Methodology - Machine learning

Year:

2016

Type of Publication:

In Proceedings

Authors:

• Sánchez-Monedero, Javier
• Pérez-Ortiz, María
• Sáez, A.
• Gutiérrez, Pedro Antonio
• Hervás-Martínez, César

Book title:

Proceedings of the 1st Workshop on Advances and Applications of Data Science & Engineering

Pages:

113-118

BibTex:

@conference{DAMA2016Melanoma, author = "Javier S{\'a}nchez-Monedero and Mar{\'i}a P{\'e}rez-Ortiz and A. S{\'a}ez and Pedro Antonio Guti{\'e}rrez and C{\'e}sar Herv{\'a}s-Mart{\'i}nez", abstract = "Melanoma is a type of cancer that develops from the pigment-containing cells known as melanocytes. Usually occurring on the skin, early detection and diagnosis is strongly related to survival rates. In the present work, we propose a system combining image analysis and machine learning for detecting melanoma presence and severity. The severity is assessed in terms of melanoma thickness, which is measured by the Breslow index. We extract 100 features considering the shape, colour, pigment network and texture of the benign and malignant lesions, tackling this problem as a five-class classification problem, where the first class represents benign lesions and the remaining four classes represent different stages of the melanoma (as measured by the Breslow index). From a machine learning point of view, this problem is a partially ordered classification task. Because of this, we propose specific machine learning models to exploit the partial order information. In this sense, we experimentally demonstrate that specifically designed models achieve better performance than a set of nominal and ordinal classifiers, considering both the imbalanced nature of the problem and the magnitude of the ordinal error in the prediction.", booktitle = "Proceedings of the 1st Workshop on Advances and Applications of Data Science {\&}amp; Engineering", pages = "113-118", title = "{M}achine learning decomposition models for partial ordering problems: {A}n application to melanoma severity classification", year = "2016", }

Abstract:

Melanoma is a type of cancer that develops from the pigment-containing cells known as melanocytes. Usually occurring on the skin, early detection and diagnosis is strongly related to survival rates. In the present work, we propose a system combining image analysis and machine learning for detecting melanoma presence and severity. The severity is assessed in terms of melanoma thickness, which is measured by the Breslow index. We extract 100 features considering the shape, colour, pigment network and texture of the benign and malignant lesions, tackling this problem as a five-class classification problem, where the first class represents benign lesions and the remaining four classes represent different stages of the melanoma (as measured by the Breslow index). From a machine learning point of view, this problem is a partially ordered classification task. Because of this, we propose specific machine learning models to exploit the partial order information. In this sense, we experimentally demonstrate that specifically designed models achieve better performance than a set of nominal and ordinal classifiers, considering both the imbalanced nature of the problem and the magnitude of the ordinal error in the prediction.