An explainable AI framework for common facial skin condition diagnosis: a design science research study
รหัสดีโอไอ
Title An explainable AI framework for common facial skin condition diagnosis: a design science research study
Creator Piyawat Setthitikun
Contributor Noptanit Chotisarn, Advisor
Publisher Thammasat University
Publication Year 2568
Keyword Explainable AI (XAI), Explanation user interface, Vision transformer, Grad-CAM++, LIME, Design science research, Clinical decision support, Dermatology, Acne, Rosacea
Abstract In dermatology, AI systems can match the diagnostic accuracy of board-certified dermatologists, yet clinical adoption remains limited because the predictionsare hard for clinicians to interpret, trust, and act on. Explainable AI (XAI) aims tomake these decisions transparent, but methods such as Grad-CAM heatmaps that areeffective in radiology may not transfer to facial dermatology, where the diagnosticcues differ. As a result, no evidence-based design guidance exists for integrating XAIinto dermatological practice.This study follows a Design Science Research methodology with threephases: literature-driven requirements analysis, prototype development, and evaluationof the constructed artifact. The prototype combines four deep learning models ofdifferent architectures, a pure Vision Transformer, two hybrid Transformers, and amodernized CNN, with three explanation methods: Grad-CAM++, Attention Rollout,and LIME. Its technical evaluation produces two results. The first is a doubly-validated evidence mechanism that retains only the regions where two independentexplanation methods agree. The second is a cross-architecture concordance analysis,which shows that models can agree on a classification while attending to entirelydifferent regions of an image, a finding that shapes how explanations should bepresented to clinicians. The study then combines these technical results with thereviewed literature to produce a Design Principles Framework: a set of nascent,evidence-based principles for integrating XAI into facial dermatology. The frameworkgives developers and healthcare decision-makers a basis for designing XAI interfacesthat support clinicians’ clinical reasoning and the adoption of AI as a decision-supporttool.
Thammasat University

บรรณานุกรม

EndNote

APA

Chicago

MLA

ดิจิตอลไฟล์

Digital File #1
DOI Smart-Search
สวัสดีค่ะ ยินดีให้บริการสอบถาม และสืบค้นข้อมูลตัวระบุวัตถุดิจิทัล (ดีโอไอ) สำนักงานการวิจัยแห่งชาติ (วช.) ค่ะ