Related Experiment Video
Updated: May 12, 2026
High Resolution 3D Imaging of the Human Pancreas Neuro-insular Network
Published on: January 29, 2018
MARTHA - Combining gaze into deep learning for fully quantitative human testicular histology analysis
Jacqueline Kockwelp1, Sabine Kliesch2, Jörg Gromoll2
1Centre of Reproductive Medicine and Andrology, University Hospital Münster, Münster, 48149, North Rhine-Westphalia, Germany; Institute for Geoinformatics, University of Münster, Münster, 48149, North Rhine-Westphalia, Germany; Faculty of Mathematics and Computer Science, University of Münster, Münster, 48149, North Rhine-Westphalia, Germany.
MARTHA, a tool combining eye tracking and deep learning, enhances pathology diagnostics. It captures expert attention to improve AI training and accuracy in analyzing whole slide images for better patient outcomes.
Area of Science:
- Computational pathology
- Biomedical image analysis
- Digital health
Background:
- Manual tissue examination is the diagnostic gold standard but lacks data for AI training.
- Pathologists' examination strategies are often lost, hindering deep learning model development.
- Current computational pathology tools struggle to integrate expert knowledge effectively.
Purpose of the Study:
- To introduce MARTHA, a novel tool integrating passive eye tracking with deep learning for enhanced diagnostics.
- To improve the efficiency and accuracy of computer-assisted diagnostics by incorporating expert attention.
- To create the largest annotated dataset for human testicular tissues.
Main Methods:
- MARTHA combines passive eye tracking with deep learning-based image analysis.
- A user-friendly graphical interface supports traditional interaction methods.
- The tool was evaluated on human testicular tissue whole slide images.
Main Results:
- MARTHA demonstrated strong performance in data interaction efficiency and semantic segmentation.
- The largest annotated dataset for human testis was generated, including over 83,000 cell nuclei.
- The tool offers valuable insights into testicular phenotypes, aiding pathologists' analyses.
Conclusions:
- MARTHA effectively integrates expert attention into AI, improving diagnostic accuracy and efficiency.
- The tool facilitates the creation of large, annotated datasets for deep learning in pathology.
- This approach supports pathologists and is a crucial step towards improving the diagnosis and treatment of male infertility.
