gms | German Medical Science

Objective: Intra-operative indocyanine green (ICG) angiography provides qualitative information on cerebral blood perfusion. For quantitative feedback, a number of challenges have to be resolved. The envisioned application is providing a near-real time feedback to the surgeon.

Methods: Video sequences of 25 ICG angiographies (21 patients), recorded over a period of eight years, were evaluated. The videos were analyzed in batch, using dedicated computer software in Java and Python, developed by the first author. The core of the computation is deconvolution of the arterial input function (AIF) from the observed tracer intensity, as in CT and MRT perfusion measurement.

For each pixel in the video, the software computes the mean transit time (MTT), time to maximum of the residue function (Tmax), regional cerebral blood flow (rCBF) and volume (rCBV). The computed values are visualized as static, color-coded images. In an intermediary step, the software also shows a graph of the computed residue function for healthy tissue.

The results were evaluated for plausibility by comparing the computed intermediary and final values with values from literature.

Results: The AIF can only be observed through vessel walls, which is an inherent challenge of the method. This was resolved by subtracting the intensity curve of healthy tissue from the arterial curve. All videos suffered from inhomogeneous illumination and had to be equalized using healthy tissue curves from different parts of the brain as references. Vibrations of the camera were also an issue in all videos, and six also suffered from strong brain pulsation. Both were compensated for using phase correlation and piece-wise affine transform, albeit in three cases without success. Due to higher time and space resolution than CT and MRT, oscillations in the deconvoluted residue function were also stronger and had to be smoothed out using a combination of sliding window and adaptive filtering. In 14 videos at least one arterial curve reached saturation (i.e. max. white value). This was mitigated by averaging over multiple curves.

A visualization of Tmax for a patient is shown in the Figure1 [Fig. 1]. Healthy tissue and arteries are bright, veins and areas with delayed flow dark.

Conclusion: Quantifying cerebral perfusion from ICG angiography is possible, but requires a skilled software operator to fine-tune the parameters. The biggest challenge for near-real time application is image stabilization, which takes 5-10 minutes per video.