Yan Tan, MSc, Simon Fraser University; Geoffrey Tien, MSc; M. Stella Atkins, PhD; Bruce Forster, MD

The speed and accuracy of placing a mouse click on an image can be improved by performing eyegaze tracking while the radiologist views an image, and then using the eyegaze position on the clickable target to dynamically alter the mouse speed towards the target.

In order for image viewing software and user interaction hardware to be valuable, it must display the images in a manner that is useful for user tasks. Pointing and clicking at targets on images are very common tasks for radiologists. When an interesting “target” is found, it may be necessary to click on that target, usually requiring mouse movement and a click. Typically, when MR and CT images have an axial and a coronal view displayed on two monitors, it is necessary to routinely move from one monitor to another for cross-referencing. Additionally, scrolling through 2D slice sequences of 3D volume data typically requires a lot of mouse movement.[1] It is easy to become fatigued and stressed after completing hundreds of such procedures, particularly when the target place to click is small, and when the distance to move is large, such as across two display monitors.[2]

The goal of this research is to improve performance of such target selection by reducing the index of difficulty identified by Fitt's Law, which states that the index of difficulty in pointing to a target is proportional to the ratio of the logarithm of distance/size.[3] Neither speeding up nor slowing down use of the mouse can change the index of difficulty, since the target size and the distance to travel are affected simultaneously with the same proportion. Thus, to reduce the index of difficulty and improve performance for pointing tasks, we need to dynamically change the mouse speed according to the distance to the target.

The difficulty in implementing this determining where the target is. Currently, several target position prediction methods have been proposed.[4,5] Many are based on information gathered from mouse movement, predicting the place (always pre-set) where the user may want to click. In our case, where radiologists want to click is unpredictable, since suspicious features can be anywhere in the image. We hypothesize that using an eye-gaze tracker to predict the target position will work well, since previous work has found that people intend to look at the place they want to click.[6] We will then dynamically adjust the mouse speed according to the distance to the target, slowing down the mouse as it reaches the target. Thus, we should simultaneously get a bigger effective target and a shorter distance in motor space, which should provide a performance gain for pointing tasks.

The pilot studies presented here illustrate the utility of the method and show the possible performance gains.

We implemented our own mouse cursor control to take advantage of dynamically adjusting the mouse Control-to-Display gain ratio, which is the ratio of the distance the mouse moves in physical (motor) space, to the corresponding distance the cursor moves on the display.[7] In these experiments, we increased the C-D ratio over the target, so the target became effectively larger (i.e., a larger movement of the physical device was required to cross the target). Since the C-D ratio was adjusted according to the distance to the target, we had to determine the position of the target on the screen. This was achieved using a Tobii 1720 eyegaze tracker. We implemented a real-time system to record the current fixation position, and the current mouse position. As the user moved the mouse toward the target, the C-D ratio was increased as the mouse approached the target under the fixation gaze.

We adjusted both the slowing-down area, and the slowing-down size of the known targets, to ascertain how much the pointing time changed, by increasing the C-D ratio dynamically. Our experimental setup is shown in Figure 1, where the mouse cursor starts near the centre of the right display, and the target appears in a random position at a certain distance from the initial position. The target appears on the left display of the Tobii monitor, which is suitable for eyegaze tracking.

Figure 1

We performed a pilot-user study to assess the impact of dynamically adjusting the C-D ratio on clicking performance on a target. Subjects are graduate students with an abundance of mouse clicking experience. For each subject, there are 15 parameter combinations, with each combination consisting of 3 target distances. For each target distance, there are four tasks with the same distance, but different (random) positions. Note that so far, we have not applied “acceleration” when the mouse is far away. This may offer more performance gain.

Figures 2 and 3 show movement time under different slowing down ratios (SR) (between 2 and 3.3 times), and slowing down area sizes (S) (between 60 and 120 pixels). Figure 2 shows the result for targets having 20 pixels width, and Figure 3 shows the results for smaller targets having 10 pixels width. In both the figures, the X axis is the target distance (pixels) and the Y axis is the mouse movement time (seconds). For each target distance, bars with different color represent different slowing down ratio and slowing down area size. In both figures, the bars labelled S(60), S(80) and S(120) mean the slowing down area width is 60, 80 or 120 pixels. The BASE (black bar) uses a constant C-D ratio corresponding to the pointer speed of 10 in Windows XP SP2.

Figure 2

Figure 3

In both figures, when the distance goes up to 1800 pixels, the performance gain can be up to 25%. Between the two sizes of targets, the smaller one (width=10 pixels) can benefit more when compared to the BASE case. These initial results show that movement time for the same task can be reduced by the appropriate values of two parameters: slowing down ratio (SR), and slowing down area size (S). Previous work has shown that the performance can also be improved under a large slowing down ratio.[8] Unfortunately, we cannot take advantage of the large slowing down ratio due to the inaccuracy of the eye-gaze tracker. If the predicted position is not the exact position of the real target, the mouse will slow down in the wrong place, which harms the performance. Thus we take the compromised scheme that uses a smaller slowing down ratio but a bigger slowing area size.

The results indicate that the performance can be improved with small slowing down ratio and a big slowing down area size, even with the inaccurate prediction of eye-gaze tracker. This is much more meaningful to the practical tasks and deployment on the radiologist stations.