November 8, 2014
Notes: Spiewak, Radoslaw
Research Support, Non-U.S. Gov’t
Acta Dermatovenerol Croat. 2013 Dec;21(4):230-5.
Author Address: Prof. Radoslaw Spiewak, MD, PhD, Department of Experimental Dermatology and Cosmetology, Faculty of Pharmacy, Jagiellonian University ul. Medyczna 9, 30-688 Krakow, Poland; email@example.com.
Reference Type: Journal Article
Record Number: 4561Author: Spyridonos, P., Gaitanis, G., Bassukas, I. D. and Tzaphlidou, M.
Title: Gray Hausdorff distance measure for medical image comparison in dermatology: Evaluation of treatment effectiveness by image similarity
Journal: Skin Res Technol
Short Title: Gray Hausdorff distance measure for medical image comparison in dermatology: Evaluation of treatment effectiveness by image similarity
Alternate Journal: Skin research and technology : official journal of International Society for Bioengineering and the Skin
ISSN: 1600-0846 (Electronic)
Accession Number: 23020792
Combined Modality Therapy
Image Processing, Computer-Assisted/*methods
Pattern Recognition, Automated/*methods
Severity of Illness Index
Abstract: INTRODUCTION: In clinical dermatology, the stabilization of the overall skin condition can be in many cases the earliest qualitative measure of the effectiveness of the therapeutic intervention. Subjective image comparisons, that offer empirical ‘qualitative’ judgments of degrees of image similarities, are traditionally employed by the involved physicians. OBJECTIVES: To quantify, by means of an image similarity metric, the degree of stabilization of an expanding skin disease, and to identify the situation of ‘no further change’ of the skin condition of the patient, providing thus the physician with an early, objective measure of the efficacy of the used therapy. METHODS: For treatment assessment, a variant of gray Hausdorff distance metric was employed to compare images of lesional skin segments of a patient, taken at different time points during a therapeutic course. Prior to image comparison, an effective preprocessing scheme was adapted to constrain wide pose and light variations. The proposed similarity algorithm was tested on raw clinical image data sets of patients diagnosed with toxic epidermal necrolysis, a life-threatening condition with rapid evolution. Fine tuning of algorithm’s parameters was optimized using Precision-Recall curves. RESULTS: Proposed image comparison method resulted in a high-degree of image similarity (about 96%) between pictures taken at second and fifth day of hospitalization. Current similarity results substantiate a significant agreement between the computer-treatment assessment, by means of image comparison, and the corresponding clinical experts’ review of skin condition. CONCLUSION: Objective evidence of ‘no further change’ situation may provide (a) intuitive clinical decision support to dermatologists in assessing aggressive skin conditions, where the timely evaluation of treatment response is of vital importance and (b) a versatile end-point measure for corresponding therapeutic clinical trials.