TY - JOUR

T1 - Stratum corneum nanotexture feature detection using deep learning and spatial analysis

T2 - a noninvasive tool for skin barrier assessment

AU - Wang, Jen Hung

AU - Pereda, Jorge

AU - Du, Ching Wen

AU - Chu, Chia Yu

AU - Christensen, Maria Oberländer

AU - Kezic, Sanja

AU - Jakasa, Ivone

AU - Thyssen, Jacob P.

AU - Satheesh, Sreeja

AU - Hwu, Edwin En Te

N1 - Publisher Copyright: © The Author(s) 2024. Published by Oxford University Press GigaScience.

PY - 2024/12/4

Y1 - 2024/12/4

N2 - Background: Corneocyte surface nanoscale topography (nanotexture) has recently emerged as a potential biomarker for inflammatory skin diseases, such as atopic dermatitis (AD). This assessment method involves quantifying circular nano-size objects (CNOs) in corneocyte nanotexture images, enabling noninvasive analysis via stratum corneum (SC) tape stripping. Current approaches for identifying CNOs rely on computer vision techniques with specific geometric criteria, resulting in inaccuracies due to the susceptibility of nano-imaging techniques to environmental noise and structural occlusion on the corneocyte. Results: This study recruited 45 AD patients and 15 healthy controls, evenly divided into 4 severity groups based on their Eczema Area and Severity Index scores. Subsequently, we collected a dataset of over 1,000 corneocyte nanotexture images using our in-house high-speed dermal atomic force microscope. This dataset was utilized to train state-of-the-art deep learning object detectors for identifying CNOs. Additionally, we implemented a kernel density estimator to analyze the spatial distribution of CNOs, excluding ineffective regions with minimal CNO occurrence, such as ridges and occlusions, thereby enhancing accuracy in density calculations. After fine-tuning, our detection model achieved an overall accuracy of 91.4% in detecting CNOs. Conclusions: By integrating deep learning object detector with spatial analysis algorithms, we developed a precise methodology for calculating CNO density, termed the Effective Corneocyte Topographical Index (ECTI). The ECTI demonstrated exceptional robustness to nano-imaging artifacts and presents substantial potential for advancing AD diagnostics by effectively distinguishing between SC samples of varying AD severity and healthy controls.

AB - Background: Corneocyte surface nanoscale topography (nanotexture) has recently emerged as a potential biomarker for inflammatory skin diseases, such as atopic dermatitis (AD). This assessment method involves quantifying circular nano-size objects (CNOs) in corneocyte nanotexture images, enabling noninvasive analysis via stratum corneum (SC) tape stripping. Current approaches for identifying CNOs rely on computer vision techniques with specific geometric criteria, resulting in inaccuracies due to the susceptibility of nano-imaging techniques to environmental noise and structural occlusion on the corneocyte. Results: This study recruited 45 AD patients and 15 healthy controls, evenly divided into 4 severity groups based on their Eczema Area and Severity Index scores. Subsequently, we collected a dataset of over 1,000 corneocyte nanotexture images using our in-house high-speed dermal atomic force microscope. This dataset was utilized to train state-of-the-art deep learning object detectors for identifying CNOs. Additionally, we implemented a kernel density estimator to analyze the spatial distribution of CNOs, excluding ineffective regions with minimal CNO occurrence, such as ridges and occlusions, thereby enhancing accuracy in density calculations. After fine-tuning, our detection model achieved an overall accuracy of 91.4% in detecting CNOs. Conclusions: By integrating deep learning object detector with spatial analysis algorithms, we developed a precise methodology for calculating CNO density, termed the Effective Corneocyte Topographical Index (ECTI). The ECTI demonstrated exceptional robustness to nano-imaging artifacts and presents substantial potential for advancing AD diagnostics by effectively distinguishing between SC samples of varying AD severity and healthy controls.

KW - atomic force microscope (AFM)

KW - atopic dermatitis (AD)

KW - corneocyte surface topography

KW - deep learning

KW - kernel density estimator (KDE)

KW - object detection

UR - http://www.scopus.com/inward/record.url?scp=85212908933&partnerID=8YFLogxK

U2 - 10.1093/gigascience/giae095

DO - 10.1093/gigascience/giae095

M3 - Article

C2 - 39657103

AN - SCOPUS:85212908933

VL - 13

JO - GIGASCIENCE

JF - GIGASCIENCE

SN - 2047-217X

M1 - giae095

ER -