Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EvidenceMoE: A Physics-Guided Mixture-of-Experts with Evidential Critics for Advancing Fluorescence Light Detection and Ranging in Scattering Media

Ismail Erbas, Ferhat Demirkiran, Karthik Swaminathan, Naigang Wang, Navid Ibtehaj Nizam, Stefan T. Radev, Kaoutar El Maghraoui, Xavier Intes, Vikas Pandey

TL;DR

FLiDAR in scattering media struggles to disentangle depth from intrinsic fluorescence lifetime due to signal convolution. We propose EvidenceMoE, a physics-guided mixture-of-experts with Evidence-Based Dirichlet Critics and a Decider for adaptive fusion, enabling end-to-end depth and lifetime estimation from raw time-resolved FLiDAR signals. The approach assigns temporally specialized experts (Early for depth, Late for lifetime, Global for context) and uses critic-derived quality scores and corrective signals to guide fusion. On realistically simulated FLiDAR data, EvidenceMoE achieves NRMS errors of 0.030 for depth and 0.074 for fluorescence lifetime, with high per-output quality scores, demonstrating robust performance and potential for real-time clinical deployment.

Abstract

Fluorescence LiDAR (FLiDAR), a Light Detection and Ranging (LiDAR) technology employed for distance and depth estimation across medical, automotive, and other fields, encounters significant computational challenges in scattering media. The complex nature of the acquired FLiDAR signal, particularly in such environments, makes isolating photon time-of-flight (related to target depth) and intrinsic fluorescence lifetime exceptionally difficult, thus limiting the effectiveness of current analytical and computational methodologies. To overcome this limitation, we present a Physics-Guided Mixture-of-Experts (MoE) framework tailored for specialized modeling of diverse temporal components. In contrast to the conventional MoE approaches our expert models are informed by underlying physics, such as the radiative transport equation governing photon propagation in scattering media. Central to our approach is EvidenceMoE, which integrates Evidence-Based Dirichlet Critics (EDCs). These critic models assess the reliability of each expert's output by providing per-expert quality scores and corrective feedback. A Decider Network then leverages this information to fuse expert predictions into a robust final estimate adaptively. We validate our method using realistically simulated Fluorescence LiDAR (FLiDAR) data for non-invasive cancer cell depth detection generated from photon transport models in tissue. Our framework demonstrates strong performance, achieving a normalized root mean squared error (NRMSE) of 0.030 for depth estimation and 0.074 for fluorescence lifetime.

EvidenceMoE: A Physics-Guided Mixture-of-Experts with Evidential Critics for Advancing Fluorescence Light Detection and Ranging in Scattering Media

TL;DR

FLiDAR in scattering media struggles to disentangle depth from intrinsic fluorescence lifetime due to signal convolution. We propose EvidenceMoE, a physics-guided mixture-of-experts with Evidence-Based Dirichlet Critics and a Decider for adaptive fusion, enabling end-to-end depth and lifetime estimation from raw time-resolved FLiDAR signals. The approach assigns temporally specialized experts (Early for depth, Late for lifetime, Global for context) and uses critic-derived quality scores and corrective signals to guide fusion. On realistically simulated FLiDAR data, EvidenceMoE achieves NRMS errors of 0.030 for depth and 0.074 for fluorescence lifetime, with high per-output quality scores, demonstrating robust performance and potential for real-time clinical deployment.

Abstract

Fluorescence LiDAR (FLiDAR), a Light Detection and Ranging (LiDAR) technology employed for distance and depth estimation across medical, automotive, and other fields, encounters significant computational challenges in scattering media. The complex nature of the acquired FLiDAR signal, particularly in such environments, makes isolating photon time-of-flight (related to target depth) and intrinsic fluorescence lifetime exceptionally difficult, thus limiting the effectiveness of current analytical and computational methodologies. To overcome this limitation, we present a Physics-Guided Mixture-of-Experts (MoE) framework tailored for specialized modeling of diverse temporal components. In contrast to the conventional MoE approaches our expert models are informed by underlying physics, such as the radiative transport equation governing photon propagation in scattering media. Central to our approach is EvidenceMoE, which integrates Evidence-Based Dirichlet Critics (EDCs). These critic models assess the reliability of each expert's output by providing per-expert quality scores and corrective feedback. A Decider Network then leverages this information to fuse expert predictions into a robust final estimate adaptively. We validate our method using realistically simulated Fluorescence LiDAR (FLiDAR) data for non-invasive cancer cell depth detection generated from photon transport models in tissue. Our framework demonstrates strong performance, achieving a normalized root mean squared error (NRMSE) of 0.030 for depth estimation and 0.074 for fluorescence lifetime.

Paper Structure

This paper contains 28 sections, 19 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Physics-Guided Inductive Biases for FLiDAR Temporal Segmentation
  4. Mixture-of-Experts (MoE)
  5. Limitations of Standard Uncertainty Quantification in Deep Learning
  6. Evidential Deep Learning
  7. Model Architecture
  8. High-Level Overview and Signal Flow
  9. Physics-Guided Mixture-of-Experts: Temporal Specialization
  10. Evidence-Based Dirichlet Critics (EDCs)
  11. Decider Network: Adaptive and Informed Fusion
  12. Phased Training Strategy
  13. Empirical Studies
  14. FliDAR data simulation in scattering media
  15. Evaluation Metrics
  16. ...and 13 more sections

Figures (4)

  • Figure 1: End-to-end EvidenceMoE Workflow for FLiDAR-based Tumor Lifetime and Depth Estimation. FLiDAR signal acquisition begins with laser interaction with a fluorescent tumor target. The resulting emitted/scattered photons are captured by a time-resolved camera, yielding a complex temporal signal. EvidenceMoE architecture processes the captured signal to ultimately yield estimations for tumor depth and fluorescence lifetime.
  • Figure 2: Overall architecture of the Physics-guided EvidenceMoE model with Evidence Critics.
  • Figure 3: Visualization of Pooled Attention Weights Over Time for Both Depth and Lifetime Experts, Computed Over a Batch Size of 512.
  • Figure 4: Performance results of EvidenceMoE model