Label-Free Intraoperative Mean-Transition-Time Image Generation Using Statistical Gating and Deep Learning

It is of paramount importance to visualize blood dynamics intraoperatively, as this enables the accurate diagnosis of intraoperative conditions and facilitates informed surgical decision-making. Indocyanine green (ICG) fluorescence imaging represents the gold standard for the assessment of blood flo...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Shi, Yan, Zhao, Denghui, Yu, Jingyi, Ni, Wei, Li, Pengcheng, Gu, Yun, Miao, Peng, Tong, Shanbao
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:It is of paramount importance to visualize blood dynamics intraoperatively, as this enables the accurate diagnosis of intraoperative conditions and facilitates informed surgical decision-making. Indocyanine green (ICG) fluorescence imaging represents the gold standard for the assessment of blood flow and the identification of vascular structures. However, it has several disadvantages, including time-consuming data acquisition, mandatory waiting periods, potential allergic reactions, and complex operations. Laser speckle contrast imaging (LSCI) provides an alternative for label-free assessment of blood flow; however, it lacks the necessary information for distinguishing arteries from veins and determining blood flow direction. Such information may be inferred from a Mean Transition Time (MTT) image derived from fluorescence imaging. In order to address these challenges, we propose the implementation of a Mixed Attention Dense UNet (MA-DenseUNet), which will be used to generate synthetic MTT images based on statistically gated deep tissue contrast and white light images. The proposed method provides clear delineation of vasculature, differentiation of arteries and veins, decoding of blood flow direction, and a reduction in imaging time by a minimum of 97.69%. This study demonstrates the potential of deep learning to optimize intraoperative optical imaging techniques, thereby providing faster and more efficient label-free surgical guidance.
DOI:10.48550/arxiv.2411.16039