Appendix

Detailed specification of used devices and software.

Surgery and supportive methods

All patients had a preoperative MRI scan (1.5 Tesla, according to the site-specific protocol, including contrast enhanced (CE) T1, T2-weighted, spectroscopy, diffusion weighed scenes).

All surgeries were aided by the StealthViz application (Medtronic Inc., Minneapolis). Blue light filtering of the neurosurgical microscope (OPMI Pentero with fluorescence kit BLU 400, Carl Zeiss, Germany) was in compliance with the requirements of the European Medical Device Directive 93/42/EEC. The excitation and observation wavelength ranges were 400-410 and 620-710 nanometres respectively. The configuration of room luminance remained constant throughout the study period. The microscope was equipped with the optional integrated digital camera MKC-500HD (Ikegami Co.; Tokyo, Japan) available from the microscope manufacturer. The following cameraspecifications were selected for the study purposes: high-definition matrix 1920x1080p, 54dB signal-to-noise ratio, sensitivity of F12 at 2000 lux/3200°K, all generated by a 1/3-inch CMOS image sensor. The pre-defined camera profiles of light sensitivity and white balance remained constant. Detailed data of spectral sensitivity and light filtering in both the microscope and camera were unobtainable from the manufacturers.

According to manufacturer’s information, the visualization of the fluorescence signal is influenced by: the illumination intensity of the light source, transmission of the optical system, total magnification, working distance, illuminated field diameter, room lighting, operating field seen either on monitor or through the eyepieces, low camera shutter, gradually elapsing fluorescence as well as perceptual non-uniformities of human eye (brochure: ZEISS OPMI® Pentero®).

StealthStation S7 navigation (Medtronic Navigation; Louisville, CO, USA) facilitated all surgeries; in three patients an intraoperative low-field MRI (PoleStar, Medtronic Navigation, Louisville, CO, USA) was used to visualize the tumour remnant.

The extent of resection, including GTR, was assessed by comparing the preoperative contrast enhancing T1-weighted tumour volume with the postoperative one. In accordance with the RANO criteria, this study included an exam between 48 and 72 hours post-op 37. The overall survival data was gathered, whereas the impact of either type of adjuvant therapy or additional surgical therapy were not evaluated.

Objectives, study design and analysis

The navigation screen was recorded in high definition throughout the surgery (Nikon D7100; Tokyo, Japan). It was synchronized with timeline of the video recorded in the operating microscope (MPEG-4 compression). The selection criteria for probing samples for the colorimetric evaluation were as follows:subjectively noticed spot demonstrating any fluorescence, circumscribed by surrounding of a different colour, regardless of its later surgical removal, probed from the movie until the evidentbrain shift has been noticed. The correspondence between contrast's enhancing in T1 and fluorescence's colour feature measures was assessed, though carries a potential bias risk due to the possible concurrent brain shift. The combined video was viewed side by side in Shotcut software v.5.2.0 (Digia Inc.; Santa Clara, CA, USA) without any digital filtering and was displayed on Dell U2410 32’’ monitor (1920x1200p, 16:10 widescreen aspect ratio, 80000:1 dynamic contrast ratio, color depth of 1.07 billion, 100% coverage of CIE L*a*b* colour gamut 7).

The fluorescent colour features of tumour remnants were compared to those of other CE regions. False - or + rates for fluorescence were not calculated as the tissue samples were not separately assessed for histopathology.In order to increase their visibility, some videos were post-processed in Shotcut using filters tailored to the obtained images. The colours of areas corresponding to missed tumour and their backgrounds were sampled twice, before and after the post-processing.

From the technical point of view, a group of various fluorescent colour features were analysed. The points of interest (samples) from a video image (both the fluorescent region and the matched backgroundinside the tumour) were probed for RGB colour model values using Colour Contrast Analyser v.2.2a (WAT Consortium, int’l). Then it was mathematically transformed to a perceptually-oriented uniform colour space CIE 1976 L*a*b* (The International Commission on Illumination)7. L* denotes lightness, a* defines red/green value and b* yellow/blue value. The difference between the shades of matched interested fluorescent region and the background determined the delta-E (ΔE) and Contrast Ratio (CR). ΔE measured the difference/distance between two colours that is calculated from their all colour coordinates (L*, a* and b*), while CR compared solely relative luminance (all colours are normalized to greyscale). The first is the principal parameter describing the difference between colours and the latterhas been widely used for internet web content assessment.

The following formulascalculated colour differences:

-ΔE =

-CR =(L1 + 0.05) / (L2 + 0.05)

where L1 and L2 denote the relative luminance of the lighter and darker colour respectively; a1, b1 and a1, b1 are particular dimensions of lighter and darker colour respectively.

CR was yielded automatically by Colour Contrast Analyser software, while ΔE as well as RGB to CIE 1976 L*a*b* transformations were calculated using a pre-defined spread sheet (Microsoft Office Excel 2010 for Windows).

(Figure 1).

Statistical analysis

The individual participant data (IPD) framework was adopted. Reports containing incomplete variables were valid for selected analyses of their complete data. Chi-square, Student’s t-test and their non-parametric equivalents tested univariate relations, whereas logistic regression was used for multivariate analysis. The area under each Receiver Operating Characteristic (ROC) curve (AUC) was calculated for significant factors. The Youden index was estimated for relevant ROCs and denoted an optimal trade-off between sensitivity and specificity. The Cox proportional hazards model explored the relationship between survival and explanatory variables. Statistica Software v.10.0 (StatSoft Inc.; Tulsa, OK, USA), and Prism (GraphPad; La Jolla, CA, USA) were used and p value <0.05 was considered statistically significant. The continuous variables were expressed as mean ± standard deviation (SD).