Bao H. Do MD, Presenter: Nothing to Disclose

John Kim MD, Abstract Co-Author: Nothing to Disclose

Joan Elizabeth Maley MD, Abstract Co-Author: Nothing to Disclose

Sandip Biswal MD, Abstract Co-Author: Nothing to Disclose

Web analytics is a method to create user profiles based on browsing behavior. We define “search pattern analytics” (SPA) as the study of radiologist search patterns through web analytics methodology. Our goal is to develop SPA tools to characterize radiologist search patterns and to measure SPA metrics in a group of radiologists.

To study radiologist reading behavior in a controlled environment, we designed a basic web-based simulator. To indirectly determine eye position during magnification, we implemented spot magnification (SM) to measure central coordinates. 15-22 radiologists (R1- subspecialist) from 2 academic institutions completed 3 cases (1 hand xray, 1 head CT/MRA, 1 ab CT) while unaware of our system. The official report (1 subspecialist) served as gold std. Training level, time to read case, slice order, and mouse & SM positions were tracked as the users studied cases. Accuracy was defined as the weighted identification of a lesion(s) without penalty for over-call. SM maps (with increasing intensity to reflect placement order) were generated for Case 1. Histograms of slices viewed (HSV) with skew calculations were generated for Case 2 and 3.

SPA statistics were tracked in real-time. SM maps: spatial and temporal data showed how users used SM to complement analysis. One pattern was a proximal to distal carpal, MCP, and DIP vertical assessment without “re-reads”; others used multiple re-reads within and among images. Total slices viewed and time required were 16.4, 18.8, 5.6 s for R1-R2, R3-R4, and staff. Histograms: Case 2: Scores for skew 75% was 2.1 vs 3.1 for mid 50% (p > 0.05). 75% R1-R2 had “extreme” skew vs 28% >R2. Case 3: Scores for skew 75% was 1.6 vs 2.0 for mid 50% (p > 0.05). 66% R1-R2 had “extreme” skew vs 28% > R2.

We have developed a novel web based method to indirectly study radiologist search patterns.

SPA requires only a browser and is an inexpensive method to indirectly study how radiologists evaluate images. SPA systems can record individualized metrics to study pattern variances.

Do, B, Kim, J, Maley, J, Biswal, S, Characterizing Radiology Search Patterns Using Web Analytics.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8016635.html