PhD ceremony Mr. I. Giotis: Advances in prototype-based learning and applications in dermatology

PhD ceremony: Mr. I. Giotis, 16.15 uur, Academiegebouw, Broerstraat 5, Groningen

Dissertation: Advances in prototype-based learning and applications in dermatology

Promotor(s): prof. N. Petkov

Faculty: Mathematics and Natural Sciences

The scope of this thesis is twofold. First, it presents two new methods in the field of prototype-based adaptive metric machine learning. One of these two methods concerns general purpose learning from vectorized data, whereas the other constitutes a novel proposal for the analysis of color texture images. Second, these methods are set in the context of dermatological diagnosis support systems. This part of the thesis focuses on the differentiation of melanoma (skin cancer) from common skin moles, a task considered one of the most challenging in the field of dermatology.

In this context chapter 2 introduces a novel algorithm, called Cluster-Based Adaptive Metric (CLAM) classification, which addresses such issues. Chapter 3 presents a novel approach for color-texture classification, called Color Image Analysis Learning Vector Quantization (CIA-LVQ) that incorporates data-driven adaptation of the system.

In order to set the aforementioned methods in the context of dermatological diagnosis, this thesis presents a quantitative study to determine the visual attributes that help dermatologists categorize lesions in different diseases. Chapter 4 presents a method that quantifies the discriminative power of the visual attributes used in dermatological diagnosis in terms of entropy. Chapter 5 integrates the above-outlined work into a computer-aided diagnosis system for melanoma, called MED-NODE, that uses simple digital images of skin lesions as input. MED-NODE's vital contributions lie first on the use of non-dermoscopic digital images which are much easier to obtain and second on the integration in the classification process of the symptoms observed by the examining physician.