LS-Net: lightweight segmentation network for dermatological epidermal segmentation in optical coherence tomography imaging

TY - JOUR

T1 - LS-Net

T2 - lightweight segmentation network for dermatological epidermal segmentation in optical coherence tomography imaging

AU - Liao, Jinpeng

AU - Zhang, Thomas

AU - Li, Chunhui

AU - Huang, Zhihong

N1 - Copyright: ©2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement.

PY - 2024/10/1

Y1 - 2024/10/1

N2 - Optical coherence tomography (OCT) can be an important tool for non-invasivedermatological evaluation, providing useful data on epidermal integrity for diagnosing skindiseases. Despite its benefits, OCT’s utility is limited by the challenges of accurate, fastepidermal segmentation due to the skin morphological diversity. To address this, we introducea lightweight segmentation network (LS-Net), a novel deep learning model that combines therobust local feature extraction abilities of Convolution Neural Network and the long-terminformation processing capabilities of Vision Transformer. LS-Net has a depth-wiseconvolutional transformer for enhanced spatial contextualization and a squeeze-and-excitationblock for feature recalibration, ensuring precise segmentation while maintaining computationalefficiency. Our network outperforms existing methods, demonstrating high segmentationaccuracy (mean Dice: 0.9624 and mean IoU: 0.9468) with significantly reduced computationaldemands (floating point operations: 1.131 G). We further validate LS-Net on our acquireddataset, showing its effectiveness in various skin sites (e.g., face, palm) under realistic clinicalconditions. This model promises to enhance the diagnostic capabilities of OCT, making it avaluable tool for dermatological practice.

AB - Optical coherence tomography (OCT) can be an important tool for non-invasivedermatological evaluation, providing useful data on epidermal integrity for diagnosing skindiseases. Despite its benefits, OCT’s utility is limited by the challenges of accurate, fastepidermal segmentation due to the skin morphological diversity. To address this, we introducea lightweight segmentation network (LS-Net), a novel deep learning model that combines therobust local feature extraction abilities of Convolution Neural Network and the long-terminformation processing capabilities of Vision Transformer. LS-Net has a depth-wiseconvolutional transformer for enhanced spatial contextualization and a squeeze-and-excitationblock for feature recalibration, ensuring precise segmentation while maintaining computationalefficiency. Our network outperforms existing methods, demonstrating high segmentationaccuracy (mean Dice: 0.9624 and mean IoU: 0.9468) with significantly reduced computationaldemands (floating point operations: 1.131 G). We further validate LS-Net on our acquireddataset, showing its effectiveness in various skin sites (e.g., face, palm) under realistic clinicalconditions. This model promises to enhance the diagnostic capabilities of OCT, making it avaluable tool for dermatological practice.

KW - Deep-learning

KW - Optical coherence tomography

KW - Skin segmentation

KW - Lightweight network

UR - https://www.scopus.com/pages/publications/85205948876

U2 - 10.1364/BOE.529662

DO - 10.1364/BOE.529662

M3 - Article

C2 - 39421780

AN - SCOPUS:85205948876

SN - 2156-7085

VL - 15

SP - 5723

EP - 5738

JO - Biomedical Optics Express

JF - Biomedical Optics Express

IS - 10

ER -