Towards Population Scale Testis Volume Segmentation in DIXON MRI

TL;DR

An evaluation of segmentation methods for testicular volume using Magnetic Resonance Imaging data from the UKBiobank shows the best model achieves a median dice score, compared to median dice score for human interrater reliability on the same dataset, enabling large-scale annotation on a population scale for the first time.

Abstract

Testis size is known to be one of the main predictors of male fertility, usually assessed in clinical workup via palpation or imaging. Despite its potential, population-level evaluation of testicular volume using imaging remains underexplored. Previous studies, limited by small and biased datasets, have demonstrated the feasibility of machine learning for testis volume segmentation. This paper presents an evaluation of segmentation methods for testicular volume using Magnet Resonance Imaging data from the UKBiobank. The best model achieves a median dice score of $0.87$, compared to median dice score of $0.83$ for human interrater reliability on the same dataset, enabling large-scale annotation on a population scale for the first time. Our overall aim is to provide a trained model, comparative baseline methods, and annotated training data to enhance accessibility and reproducibility in testis MRI segmentation research.

Figures (5)

• Figure 1: Sketch of UKBiobank DIXON measurements. From the 1.1 meters hight of measurement in the UKBiobank (a), a total of six overlapping imaging windows are extracted (b).
• Figure 2: Sample annotation screenshots from UKBiobank DIXON MRI on the water channel of the fifth imaging window. (a) and (b) are showing MRI scans of physiological testes. (c) shows a hydrocele testis. Reproduced by kind permission of UKBiobank.
• Figure 3: Split of datasets. From the 22,522 initially obtained samples, we annotate 350 randomly selected subjects. Those are split into training and test samples. A subset of 12 images of the testset is then used to calculate interrater agreement. The trainingset is used for 5-Fold Cross-Validation. Final Model is trained on all training subjects (313 subjects) and evaluated on the testset. Final inference is done on 22,172 images.
• Figure 4: Multiple annotations for one example image. Annotation from main annotator red, radiologist annotation white. Annotation is viewed as volume in one example view on sagittal, coronal and axial slices.
• Figure 5: Histogram of calculated volumes for 22,149 unlabeled DIXON MR images from UKBiobank. Count in absolute values and volume in ml.