Visualization of superficial vein dynamics in dorsal hand by near-infrared imaging in response to elevated local temperature

Abstract

Significance: Dry or moist skin-contact thermal stimulation for vein puncture (VP) and vein cannulation (VC) may not be feasible for sensitive skin. For a damaged, burned, or dark skin, near-infrared (NIR) imaging is preferred to visualize a vein. Postprocessing of NIR images is always required because the skin is a reflective material and veins need segmentation for quantitative analysis.

Aim: Our pilot study aims to observe the effect of noncontact local heating on the superficial metacarpal veins in the dorsal surface of the hand and to visualize vein dynamics using an NIR imaging system.

Approach: Our experiment consists of studies A and B at two ambient temperatures, 19°C and 25°C. A simple reflection-based NIR imaging system was installed to acquire sequential vein images for 5 min before and after applying 10 min of radiant thermal stimulation. To measure the vein diameter (VD), we trained a convolutional neural network (CNN) on sequential raw images to predict vein-segmentation masks as output images. Later these masked images were postprocessed for the VD measurements.

Results: The average VD was significantly increased after thermal stimulation in study A. The maximum increments in VD were 39.3% and 9.19%, 1 min after thermal stimulation in studies A and B, respectively. Both the VD and skin temperature (Tskin) follow negative exponentials in time, and the VD is proportional to Tskin. A multiple linear-regression model was made to predict the final VD. A significant difference was observed in the change of the VD.

Conclusions: NIR imaging with CNN can be used for quantitative analyses of vein dynamics. This finding can be further extended to develop real-time, image-guided medical devices by integrating them with a radiant heater and to assist medical practitioners in achieving high success rates for VP or VC.

Keywords: near-infrared imaging; skin temperature; vasomotor response; vein cannulation; vein diameter; vein puncture.

Fig. 1

(a) Optimal position of the heat source to achieve a target temperature was determined by measuring the temperature at different distances from an IR radiant lamp. The inset shows a schematic for measurement of temperature using a thermocouple and a thermal camera. (b) Procedure for human subject experiment.

Fig. 2

Acquisition and alignment of images at the experiment. (a) A schematic of the experimental setup for NIR image acquisition. (b) Image sequence stored in the memory for further analysis. (c) The inset in (b) as ROI is cropped and showed for geometrical correction. Upper row represents the original cropped raw images and lower row shows the geometrically corrected images. A square dashed box is drawn on the raw image by considering the dark mark at its center. Scale bar = 10 mm. (d) Quantitative analysis of the original cropped and corrected image are shown by a normalized gray value inside the dashed box centering the dark mark as shown in (c).

Fig. 3

Quantitative measurement of VD. (a) Vein existing ROI is cropped from the raw image for processing. Scale bar = 10 mm. (b) Raw image was further processed for VD measurement. (c) A quantitative measurement of an arbitrary vein is scanned in 1D position to emphasize the reason for raw image processing. The blue and red lines in the plot represent gray values of the raw and processed images, respectively. Red color line shows a significant difference from the blue line, which are used to clearly determine VD. (d) However, 1D scan gray value measurement could be confusing as the vein is not symmetric in size for its whole length. A random area from the target vein is selected to measure its average. (e) The dark red, green, and orange lines in the plot are the 1D scan gray values, respectively, for 1, 2, and 3, as shown in (d). The VD is measured by determining FWHM of the Gaussian curve. (f) A quantitative analysis is shown in the dark and shaded bar plot for VD measurement in single line scan and area scan, respectively.

Fig. 4

A repeatability study ($n=2$) of 2D vein dynamics of an individual at local heat exposure in study A condition. (a) A small region of the time sequence image is further cropped to magnify the vein dynamics due to local heat stimulation. The veins are outlined by red dashed lines in the time sequence images, where the control represents the vein image before heating and the change of the VD can be observed at $t=1$, 3, and 5 min after 10 min of local heat exposure. (b) Mean VD was measured at each time point from inset images in (a). A significant difference of VD can be seen from the bar chart. This significant difference was further justified by Student’s $t$-test ($p<0.001$). ⋆ indicates the significant difference ($p\le 0.001$) between mean VDs measured before and after local heat exposure. (c) Mean $T_{\text{skin}}$ is plotted at $t=1$, 3, and 5 min after 10-min thermal stimulation is off and compared with the control.

Fig. 5

Before and after thermal stimulation effect on $T_{\text{skin}}$ and VD. (a) and (b) Values are shown as the average of the participants’ $T_{\text{skin}}$ at studies A and B ($n=10$). After the radiation heat exposure, there was a significant difference observed for $T_{\text{skin}}$ as shown in the bar chart. (c) and (d) Vein dynamics observed from its percent of dilation for different skin types. The vein was dilated after heat exposure to the dorsal hand surface. The standard deviation of VD was big as each individual has a different percentage of the dilation from their base diameter. However, the individual change of each person has a significant difference from their base value. (e) and (f) Time average of VD change shows a significant difference between two different races as confirmed by Student’s $t$-test. * and ** indicate the significant difference of $p$-value at $p\le 0.01$ and $p\le 0.05$, respectively.

Fig. 6

(a) and (b) Individual data points of $T_{\text{skin}}$ and VD for 5 min after thermal stimulation off. (c) A scatter plot that shows the VD change with change of $T_{\text{skin}}$ regardless of the vein size and race. (d) Mean $T_{\text{skin}}$ and VD exponentially decrease with time after removing the thermal stimulation in study A condition ($n=10$). (e) Linear correlation between the $T_{\text{skin}}$ and VD is shown, and the $R^2$ value indicates a good agreement between them.

Fig. 7

(a) A multilinear-regression model was made and validated with scattered plot as shown where $R^2=0.69$. (b) Bland–Altman plot analysis of the prediction model shows $\pm 0.6\mathrm{mm}$ prediction variance form the measured value.