Near-real-time Mueller polarimetric image processing for neurosurgical intervention

Abstract

Purpose: Wide-field imaging Mueller polarimetry is a revolutionary, label-free, and non-invasive modality for computer-aided intervention; in neurosurgery, it aims to provide visual feedback of white matter fibre bundle orientation from derived parameters. Conventionally, robust polarimetric parameters are estimated after averaging multiple measurements of intensity for each pair of probing and detected polarised light. Long multi-shot averaging, however, is not compatible with real-time in vivo imaging, and the current performance of polarimetric data processing hinders the translation to clinical practice.

Methods: A learning-based denoising framework is tailored for fast, single-shot, noisy acquisitions of polarimetric intensities. Also, performance-optimised image processing tools are devised for the derivation of clinically relevant parameters. The combination recovers accurate polarimetric parameters from fast acquisitions with near-real-time performance, under the assumption of pseudo-Gaussian polarimetric acquisition noise.

Results: The denoising framework is trained, validated, and tested on experimental data comprising tumour-free and diseased human brain samples in different conditions. Accuracy and image quality indices showed significant ( $p<0.05$ ) improvements on testing data for a fast single-pass denoising versus the state-of-the-art and high polarimetric image quality standards. The computational time is reported for the end-to-end processing.

Conclusion: The end-to-end image processing achieved real-time performance for a localised field of view ( $\approx 6.5{\text{mm}}^2$ ). The denoised polarimetric intensities produced visibly clear directional patterns of neuronal fibre tracts in line with reference polarimetric image quality standards; directional disruption was kept in case of neoplastic lesions. The presented advances pave the way towards feasible oncological neurosurgical translations of novel, label-free, interventional feedback.

Keywords: AI; Mueller polarimetric imaging; Neurosurgery; Real-time denoising.

Fig. 1

Denoising diffusion model. a Forward and reverse diffusion as in [18]. Degraded states ${\mathbf{x}}_t$ of the forward diffusion (black arrow) in the modelled Markov chain of T time-points. Inferential sampling of the reverse diffusion (blue arrow) over the parametrised distribution $p_{\theta }({\mathbf{x}}_{t-1}|{\mathbf{x}}_t)$. b Schematic diagram of the AI-based denoising polarimetric framework. PDDN builds on a time-point recursive U-Net for the reverse diffusion, as in [19]

Fig. 2

Wide-field MPI system in reflection configuration. a Schematic: light source, polarisation state generator $\mathbf{G}$, biological sample, polarisation state analyser $\mathbf{A}$, and detecting camera. b Our instrumentation. c Acquired polarisation states intensity image $\mathbf{I}$. d Derived full Mueller matrix $\mathbf{M}$. e Lu-Chipman decomposition: ${\mathbf{M}}_{\mathrm{\Delta }}$, ${\mathbf{M}}_R$, and ${\mathbf{M}}_D$ matrices. f Derived scalar parameters: D, $\mathrm{\Delta }$, R, and $\phi$. Enhanced contrast of $\mathbf{M}$ with a sigmoid mapping

Fig. 3

Denoising polarimetric intensities of human brain tissues: gallery of polarimetric instances. (top) Tumour-free sample of the testing set. ${\text{I}}_{11}$ component shown with the evaluation ROI contour. (bottom) Details of polarimetric parameters in a centre-cropped area. Images in each row have the same range of values as in the colour-bar. R and $\phi$ reported in degrees

Fig. 4

Denoising polarimetric intensities of human brain tissues: gallery of polarimetric instances. (top) Neoplastic lesion of the testing set. ${\text{I}}_{11}$ component shown with the evaluation ROI contour. (bottom) Details of polarimetric parameters in a centre-cropped area. Images in each row have the same range of values as in the colour-bar. R and $\phi$ reported in degrees

Fig. 5

Azimuth $\phi$ variations and distributions: intensity and circular standard deviation csd($\phi$) in degrees. Low values of csd($\phi$) for homogeneous directional patterns, whereas high values for high disruption of fibres orientation or change of directional patterns. (top) Tumour-free sample: variability of fibres orientations, csd distributions in annotated white (WM) and grey (GM) matter. (bottom) Neoplastic lesion: variability of fibres orientations in diseased white matter, csd distributions in annotated tumour centre (TC) and infiltration (TI). High rejection of background noise, and visually comparable directional patterns of the fibres after denoising, similarly to high-quality image standards. Boxplots: consistent csd drop in PDDN as in HQ and SHQ. Better PDDN separation in tumour areas

Fig. 6

Denoised polarimetric tractography: tumour-free brain sample. LQ and denoised azimuth with recursive PDDN$_{}^{+}$. MPI PTF: parameters mapped into an ellipsoidal model for tractography. Colours code for trace of ellipsoids eigenvalues. Fibres in seeding regions as in neurosurgical probing