Table 2.

Pathologist viewing behaviors' definitions.

Viewing behavior	Definition	Equation
Total interpretation time	Using the time stamp (TS) of each viewport (vi), we calculated the duration (d) of each viewport being viewed. Total interpretation time (T) is calculated by summing the durations of all the viewports.	$d\left(v_i\right)=\mathit{TS}\left(v_{i+1}\right)-\mathit{TS}\left(v_i\right)$ $T={\sum }_{v_i=1}^{v_n}d\left(v_i\right)$
Average zoom level	The average zoom level (avg) used during an interpretation.	$\mathit{avg}=\frac{{\sum }_{v_{i=1}}^{v_n}\mathit{zoom}\left(v_i\right)}{n}$
Maximum zoom level	The maximum zoom level (mx) used during an interpretation.	$\mathit{mx}=max\left\{\mathit{zoom}\left(v_i\right):i=1..n\right\}$
Zoom level variance	The zoom level variance (var) during an interpretation.	$\mathit{var}=\frac{{\sum }_{v_{i=1}}^{v_n}{\left(\mathit{zoom}\left(v_i\right)-\mathit{avg}\right)}^2}{n-1}$
Magnification percentage	Magnification percentage (MP) is calculated based on the number of times a pathologist zooms in consecutively. This variable captures how deeply and frequently a pathologist zooms while interpreting a case. We count the number of viewports that are associated with consecutive zoom-in behavior (m). A consecutive zoom-in is a sequence of viewports where the zoom level of each viewport is greater than its previous viewport in the sequence. Magnification percentage calculates the proportion of viewports associated with this behavior.	$m\left(v_i\right)=\left\{\begin{array}{l}1,\mathit{zoom}\left(v_{i+1}\right)\ge \mathit{zoom}\left(v_i\right)\\ 0,\mathit{otherwise}\end{array}\right)$ $\mathit{MP}=\frac{{\sum }_{v_{i=1}}^{v_n}m\left(v_i\right)}{n}$
ROI time percentage	ROI time percentage (RTP) measures the amount of time a pathologist spends viewing regions that experts marked as ROI. When a pathologist's viewport intersects with the consensus ROI by 40% or more, we consider that the pathologist is viewing the consensus ROI (r). However, to ensure that the pathologist is actually attending to the ROI, we apply a size constraint. We exclude cases where the ratio of the ROI area to the viewport area is smaller than 10%. This way we make sure that a viewport intersects with a large area of the ROI and this intersection covers the most parts of the viewport. ROI time percentage calculates the proportion of interpretation time spent viewing such regions. Pathologists were not informed about the consensus ROI at the time of interpretation, spending more time on such regions means that they independently identified the region as important.	$r\left(v_i\right)=\left\{\begin{array}{ll}1,& \frac{\mathit{area}\left(\mathit{intersect}\left(v_i,\mathit{ROI}\right)\right)}{\text{area}\left(\mathrm{ROI}\right)}>0.4\\ & \mathit{And}\\ & \frac{\mathit{area}\left(\mathit{ROI}\right)}{\mathit{area}\left(v_i\right)}>0.1\\ 0,& \mathit{otherwise}\end{array}\right)$ $\mathit{RTP}=\frac{{\sum }_{v_{i=1}}^{v_n}r\left(v_i\right)}{n}$
Scanning percentage	Scanning percentage (SP) is defined similar to the E. Mercan et. al12 study for digital breast pathology. This variable provides information regarding zoom level changes in consecutive log entries, regardless of zoom level itself. We count the number of viewports that are associated with consecutive zoom in behavior (s). Scanning percentage calculates the proportion of viewports associated with panning around the image with a fixed zoom level.	$s\left(v_i\right)=\left\{\begin{array}{l}1,\mathit{zoom}\left(v_{i+1}\right)==\mathit{zoom}\left(v_i\right)\\ 0,\mathit{otherwise}\end{array}\right)$ $\mathit{SP}=\frac{{\sum }_{v_{i=1}}^{v_n}s\left(v_i\right)}{n}$