@PHDTHESIS\{IMM2003-02454,
    author       = "S. Sigurdsson",
    title        = "A Probabilistic Framework for Detection of Skin Cancer by Raman Spectra",
    year         = "2003",
    pages        = "202",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    note         = "Supervisor: Lars Kai Hansen",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/2454-full.html",
    abstract     = "This Ph.D. thesis focuses on objective methods for diagnosing skin cancer from Raman spectra. A method for suppressing background noise and dimension reduction in Raman spectra is suggested. A robust Bayesian framework for training a neural network is proposed, including an overfit control and outlier framework. Finally a visualization scheme for extracting important features from the trained neural network classifier based on sensitivity analysis is defined.

The performance on two types of skin cancer showed that 97.9\% of basal cell carcinoma were identified correctly and 85.5\% of malignant melanoma. The neural network classifier visualization showed that frequency bands, previously identified by visual inspection of Raman spectra by medical experts, were considered important for classification. Moreover, frequency band not previously used for skin lesion classification were identified. These identified important features are shown to originate from molecular structure changes in lipids and proteins.

While the theme of this dissertation is skin cancer diagnosis from Raman spectra, the dimension reduction and the neural network classifier can be applied in general to other types of pattern recognition problems.

In Danish:

Denne Ph.d. afhandling fokuserer p{\aa} a objektive metoder til diagnosering af hudkr{\ae}ft fra Raman spektra. En metode til d{\ae}mpning af hud fluorisence og dimensions reduktion af Raman spektra er foresl{\aa}et. En robust Bayesiansk fremgangsm{\aa}de til tr{\ae}ning af neural netv{\ae}rk er foresl{\aa}et, som indeholder outlier kontrol og overfitting h{\aa}ndtering. Endelig, er der defineret en visualisering metode af vigtige features fra det tr{\ae}nede neural netv{\ae}rk, baseret p{\aa} indgangs/ udgangs f{\o}lsomheds analyse.

Diagnose resultater for det neurale netv{\ae}rk for to typer hudkr{\ae}ft viser, at 97.9\% af basal cell carcinoma og 85.5\% af malignant melanoma er korrekt klassificeret. Visualisering viser at frekvensb{\aa}nd i Raman spektra, som hudl{\ae}ger havde identificeret som vigtige, ogs{\aa} blev identificeret som vigtige af det neurale netv{\ae}rk. Endvidere, finder det neurale netv{\ae}rk frekvens b{\aa}nd som ikke f{\o}r er brugt til diagnosering af hudkr{\ae}ft. Disse vigtige frekvensb{\aa}nd stammer fra forskel i molekyle struktur i lipider og proteiner.

Selv om temaet for denne afhandling er hudkr{\ae}ft diagnosering fra Raman spektra, kan dimensions reduceringen og det neurale netv{\ae}rk bruges til andre m{\o}nster genkendelses problemer uden videre tiltag."
}