Aristotle University Of Thessaloniki (AUTH)
Στατιστικά μοντέλα φασματογραφίας ηλεκτρικής εμπέδησης και φωτονικής σε επαληθευμένα δερματολογικά διαγνωστικά δεδομένα
Abstract
dc:descriptionAllergic contact dermatitis (ACD) is one of the most common inflammatory skin disorders, with a considerable prevalence and a substantial impact on patients’ quality of life. Its diagnosis is based on epicutaneous patch testing, which represents the reference standard for the identification of causative allergens. However, interpretation of patch test reactions relies on the International Contact Dermatitis Research Group (ICDRG) scoring system, an established clinical tool which, although widely used, is based on clinical judgment and may present inter-observer variability, particularly in borderline cases. Within the framework of the present doctoral dissertation, the use of non-invasive biophysical and imaging-based methods for the objective evaluation of skin reactions was investigated, with particular emphasis on enhancing patch test assessment through computational approaches. In an initial phase, the application of Electrical Impedance Spectroscopy (EIS) for the discrimination of skin lesions was explored. Although the method demonstrated diagnostic potential, limitations regarding diagnostic accuracy were also identified, mainly due to limited specificity and moderate discriminative performance (AUC ~0.60–0.67). These limitations, together with issues related to measurement reproducibility, led to the methodological evolution of the dissertation toward more robust quantitative imaging approaches. The main study was based on a prospective design and included 194 patients with suspected ACD (4,477 observations). The analysis focused on the utilization of data derived from the Antera 3D multispectral imaging system and the development of machine learning models using structured features associated with erythema, chromatic alterations, and skin morphology. The results demonstrated that these features can effectively discriminate positive from negative reactions, achieving high discriminative performance (AUC = 0.861). Incorporation of the temporal evolution of reactions between 48 and 72 hours further improved model performance (AUC = 0.902), highlighting the critical role of inflammatory dynamics in diagnosis. In addition, class imbalance handling and threshold optimization enabled high sensitivity levels (up to 94.9% under screening conditions), substantially reducing false-negative results. Analysis of the findings demonstrated that the most important predictive factors were primarily associated with erythema-related indices and color parameters, a finding consistent with clinical practice and confirmed by both SHAP analysis and statistical evaluation. Furthermore, a statistically significant association was identified between the allergen 2-hydroxyethyl methacrylate (HEMA) and occupational dermatitis, highlighting the potential of computational approaches to reveal clinically meaningful patterns. Overall, the findings of this dissertation suggest that combining quantitative parameters derived from Antera 3D imaging with machine learning methods may enhance objectivity and diagnostic consistency in the evaluation of patch tests. The proposed approach is not intended to replace clinical judgement, but rather to complement established clinical assessment, particularly in cases where interpretation is challenging or borderline. These findings may represent an important first step towards the development of clinical decision support systems to support clinical practice.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Gylou, Sotiria
- Γυλού, Σωτηρία
Subjects
dc:subject × 14- Φασματογραφία ηλεκτρικής εμπέδησης
- Αλλεργική δερματίτιδα εξ επαφής
- Πολυφασματική απεικόνιση
- Σύστημα υποστήριξης κλινικών αποφάσεων
- Electrical Impedance Spectroscopy
- Αllergic contact dermatitis
- Μultispectral imaging
- Clinical decision support system
- Επιστήμες Υγείας
- Ιατρική και Επιστήμες Υγείας
- Ιατρική Πληροφορική
- Health Sciences
- Medical and Health Sciences
- Medical Informatics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/61758