| Creators: |
Hieber, Daniel and Jungbäck, Nicola and Holl, Felix and Märkl, Bruno and Pryss, Rüdiger and Schobel, Johannes and Liesche-Starnecker, Friederike |
| Title: |
Using Eye-Tracking to find Differences in the Analysis ofWhole-Slide Images Between Physicians and Machine Learning Models –A Study Design |
| Item Type: |
Conference or Workshop Item |
| Event Title: |
(Meeting Abstracts of the) 68th Annual Meeting of the German Society of Neuropathology and Neuroanatomy (DGNN) |
| Event Location: |
Regensburg, Germany |
| Event Dates: |
12.-14. September 2024 |
| Projects: |
DigiHealth, NAP |
| Page Range: |
p. 45 |
| Additional Information: |
Open Access | Free Neuropathology 5:19 (2024), p.45 |
| Date: |
2024 |
| Divisions: |
Gesundheitsmanagement |
| Abstract (ENG): |
Neuropathologists are experts in their field, quickly detecting and classifying tumors in Whole-Slide Images (WSIs). However, it is often unclear what features they focus on and which aspects they rely on for their decision. Machine Learning (ML) is also increasingly used in pathology, but its decision-making process is similarly opaque. To better understand expert analysis and evaluate ML reasoning, an eye-tracking study will compare detectionaccuracy and Region of Interest (RoI) selection among neuropathologists, pathologists in general, medicalstudents, and ML models.The study involves experts (neuropathologists and pathologists with high experience in brain tumor diagnostics), intermediates (pathologists without any particular expertise in brain tumor diagnosis), and beginners (students, physicians from other domains) analyzing hematoxylin-and-eosin (HE) stained sections of 50 Glioblastoma (GBM) WSIs. Participants must determine whether neoplastic tissue is present in the section and, if so, mark the most relevant regions for their analysis. The EyeLogic LogicOne eye-tracking device records their gaze during thisprocess.Post-study, gaze data from different participant groups will be compared with ML probability masks to identify differences and similarities between human and artificial analysts. The study will also correlate selected RoIs with ML probabilities and gaze duration to see if the most critical areas are those most viewed by humans and deemed most important by ML.An eye-tracking study is planned for October 2024, comparing the analysis of HE-stained WSIs of GBMs between human analysts and ML models. The study is currently in the recruitment phase, with results expected by end ofthe year.Meeting Abstract |
| Forthcoming: |
No |
| Language: |
English |
| Link eMedia: |
Download |
| Citation: |
Hieber, Daniel and Jungbäck, Nicola and Holl, Felix and Märkl, Bruno and Pryss, Rüdiger and Schobel, Johannes and Liesche-Starnecker, Friederike
(2024)
Using Eye-Tracking to find Differences in the Analysis ofWhole-Slide Images Between Physicians and Machine Learning Models –A Study Design.
In: (Meeting Abstracts of the) 68th Annual Meeting of the German Society of Neuropathology and Neuroanatomy (DGNN), 12.-14. September 2024, Regensburg, Germany, p. 45.
|
