PhD Scientific Days 2025

Poster Session III. - K: Theoretical and Translational Medicine

Quest for a Clinically Relevant Medical Image Segmentation Metric: the Definition and Implementation of Medical Similarity Index

Előadó neve

Dr. Fazekas Szuzina

Neptun code

JZINDU

Előadó munkahelye

Semmelweis University Medical Imaging Centre

Előadó telefonszáma

+36300981870

Előadó e-mail címe

szuzina1997@gmail.com

Az előadás címe

Quest for a Clinically Relevant Medical Image Segmentation Metric: the Definition and Implementation of Medical Similarity Index

Szerző(k) neve és munkahelye

Dr. Szuzina Fazekas¹, Zsolt Vizi²

1: Semmelweis University Medical Imaging Centre
2: University of Szeged, Bolyai Institute

Bemutatás módja

Poszter

Szekció

Poster Session III. - K: Theoretical and Translational Medicine

Language of the presentation

English

Preferred session

Theoretical and Translational Medicine

Összefoglaló szövege

Introduction: In the field of radiology and radiotherapy, accurate delineation of different organs plays crucial role in both diagnostics and therapeutics. While the gold standard remains expert-driven manual segmentation, many machine learning-based automatic segmentation methods are emerging. The evaluation of these methods mainly relies on traditional metrics which fail to adapt to different clinical applications. Thus, there is an understandable need for a clinically meaningful, reproducible assessment of autocontouring systems.
Aims: This study aims to develop and implement a clinically relevant segmentation metric that can be adapted to different medical imaging applications.
Method: The reference contour was considered the gold standard segmentation, the agreement of a test contour to the reference contour was quantified. Based on bidirectional local distance, the points of the test contour were paired to points of the reference contour. After correcting for the distance between the test and reference center of mass, the Euclidean distance was calculated between the paired points, and a score was given to each test point. The overall medical similarity index was calculated as the average scores across all the test points. The fine-tuning of the user-defined hyperparameters was demonstrated with an open-access anatomic prostate segmentation MRI dataset. We trained an nnUNet neural network for segmentation, and manually selected six test cases (two easy, two moderate and two hard cases) for evaluation.
Results: An easy-to-use, sustainable image processing pipeline was created using Python. The algorithm can handle multislice images with multiple masks per slice. Additionally, a mask splitting algorithm is also provided for the separation of concave masks. The clinical relevance and adaptability is demonstrated by prostate anatomic segmentations.
Conclusion: A novel segmentation evaluation metric was implemented, and an open-access image processing pipeline was also provided, which can be easily used for automatic measurement of clinical relevance of medical image segmentation. The pipeline enables the calculation of MSI and traditional segmentation metrics and fine-tuning for clinical use. This tool offers a reproducible and adaptable framework for evaluating autocontouring systems in medical imaging.
Funding: Gedeon Richter Excellence PhD Scholarship

University

Semmelweis University

Supervisor

Zsolt Vizi

Publication of my abstract

I give consent to the publication of my abstract on the website of the congress.

Kind

Szabad

Status

elfogadva

Accepted presentation method

poszter

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

7239

Start

14:30

End

14:36