Simon Pezold

Simon Pezold, Jérôme A. Kurylec, Jan S. Liechti et al.

We investigate how both the adaptation of a generic foundation model via transfer learning and the integration of complementary modalities from the operating room (OR) can support surgical data science. To this end, we use V-JEPA as the single-modality foundation of a multimodal model for minimally invasive surgery support. We analyze how the model's downstream performance can benefit (a) from finetuning on unlabeled surgical video data and (b) from providing additional time-resolved data streams from the OR in a multimodal setup. In an in-house dataset of liver surgery videos, we analyze the tasks of predicting hospital length of stay and postoperative complications. In videos of the public HeiCo dataset, we analyze the task of surgical phase recognition. As a baseline, we apply pretrained V-JEPA to all tasks. We then finetune it on unlabeled, held-out videos to investigate its change in performance after domain adaptation. Following the idea of modular decision support networks, we integrate additional data streams from the OR by training a separate encoder to form a shared representation space with V-JEPA's embeddings. Our experiments show that finetuning on domain-specific data increases model performance. On the in-house data, integrating additional time-resolved data likewise benefits the model. On the HeiCo data, accuracy of the pretrained video-only, single-modality baseline setup is on par with the top-performing submissions of the EndoVis2017 challenge, while finetuning on domain-specific data increases accuracy further. Our results thus demonstrate how surgical data science can leverage public, generic foundation models. Likewise, they indicate the potential of domain adaptation and of integrating suitable complementary data streams from the OR. To support further research, we release our code and model weights at https://github.com/DigitalSurgeryLab-Basel/ML-CDS-2025.

Hugo J. Kuijf, J. Matthijs Biesbroek, Jeroen de Bresser et al.

Quantification of cerebral white matter hyperintensities (WMH) of presumed vascular origin is of key importance in many neurological research studies. Currently, measurements are often still obtained from manual segmentations on brain MR images, which is a laborious procedure. Automatic WMH segmentation methods exist, but a standardized comparison of the performance of such methods is lacking. We organized a scientific challenge, in which developers could evaluate their method on a standardized multi-center/-scanner image dataset, giving an objective comparison: the WMH Segmentation Challenge (https://wmh.isi.uu.nl/). Sixty T1+FLAIR images from three MR scanners were released with manual WMH segmentations for training. A test set of 110 images from five MR scanners was used for evaluation. Segmentation methods had to be containerized and submitted to the challenge organizers. Five evaluation metrics were used to rank the methods: (1) Dice similarity coefficient, (2) modified Hausdorff distance (95th percentile), (3) absolute log-transformed volume difference, (4) sensitivity for detecting individual lesions, and (5) F1-score for individual lesions. Additionally, methods were ranked on their inter-scanner robustness. Twenty participants submitted their method for evaluation. This paper provides a detailed analysis of the results. In brief, there is a cluster of four methods that rank significantly better than the other methods, with one clear winner. The inter-scanner robustness ranking shows that not all methods generalize to unseen scanners. The challenge remains open for future submissions and provides a public platform for method evaluation.

Spyridon Bakas, Mauricio Reyes, Andras Jakab et al.

Gliomas are the most common primary brain malignancies, with different degrees of aggressiveness, variable prognosis and various heterogeneous histologic sub-regions, i.e., peritumoral edematous/invaded tissue, necrotic core, active and non-enhancing core. This intrinsic heterogeneity is also portrayed in their radio-phenotype, as their sub-regions are depicted by varying intensity profiles disseminated across multi-parametric magnetic resonance imaging (mpMRI) scans, reflecting varying biological properties. Their heterogeneous shape, extent, and location are some of the factors that make these tumors difficult to resect, and in some cases inoperable. The amount of resected tumor is a factor also considered in longitudinal scans, when evaluating the apparent tumor for potential diagnosis of progression. Furthermore, there is mounting evidence that accurate segmentation of the various tumor sub-regions can offer the basis for quantitative image analysis towards prediction of patient overall survival. This study assesses the state-of-the-art machine learning (ML) methods used for brain tumor image analysis in mpMRI scans, during the last seven instances of the International Brain Tumor Segmentation (BraTS) challenge, i.e., 2012-2018. Specifically, we focus on i) evaluating segmentations of the various glioma sub-regions in pre-operative mpMRI scans, ii) assessing potential tumor progression by virtue of longitudinal growth of tumor sub-regions, beyond use of the RECIST/RANO criteria, and iii) predicting the overall survival from pre-operative mpMRI scans of patients that underwent gross total resection. Finally, we investigate the challenge of identifying the best ML algorithms for each of these tasks, considering that apart from being diverse on each instance of the challenge, the multi-institutional mpMRI BraTS dataset has also been a continuously evolving/growing dataset.

Antal Horvath, Charidimos Tsagkas, Simon Andermatt et al.

The small butterfly shaped structure of spinal cord (SC) gray matter (GM) is challenging to image and to delinate from its surrounding white matter (WM). Segmenting GM is up to a point a trade-off between accuracy and precision. We propose a new pipeline for GM-WM magnetic resonance (MR) image acquisition and segmentation. We report superior results as compared to the ones recently reported in the SC GM segmentation challenge and show even better results using the averaged magnetization inversion recovery acquisitions (AMIRA) sequence. Scan-rescan experiments with the AMIRA sequence show high reproducibility in terms of Dice coefficient, Hausdorff distance and relative standard deviation. We use a recurrent neural network (RNN) with multi-dimensional gated recurrent units (MD-GRU) to train segmentation models on the AMIRA dataset of 855 slices. We added a generalized dice loss to the cross entropy loss that MD-GRU uses and were able to improve the results.

Simon Andermatt, Antal Horváth, Simon Pezold et al.

Fully supervised segmentation methods require a large training cohort of already segmented images, providing information at the pixel level of each image. We present a method to automatically segment and model pathologies in medical images, trained solely on data labelled on the image level as either healthy or containing a visual defect. We base our method on CycleGAN, an image-to-image translation technique, to translate images between the domains of healthy and pathological images. We extend the core idea with two key contributions. Implementing the generators as residual generators allows us to explicitly model the segmentation of the pathology. Realizing the translation from the healthy to the pathological domain using a variational autoencoder allows us to specify one representation of the pathology, as this transformation is otherwise not unique. Our model hence not only allows us to create pixelwise semantic segmentations, it is also able to create inpaintings for the segmentations to render the pathological image healthy. Furthermore, we can draw new unseen pathology samples from this model based on the distribution in the data. We show quantitatively, that our method is able to segment pathologies with a surprising accuracy being only slightly inferior to a state-of-the-art fully supervised method, although the latter has per-pixel rather than per-image training information. Moreover, we show qualitative results of both the segmentations and inpaintings. Our findings motivate further research into weakly-supervised segmentation using image level annotations, allowing for faster and cheaper acquisition of training data without a large sacrifice in segmentation accuracy.

Simon Andermatt, Simon Pezold, Michael Amann et al.

We present an automated method for localizing an anatomical landmark in three-dimensional medical images. The method combines two recurrent neural networks in a coarse-to-fine approach: The first network determines a candidate neighborhood by analyzing the complete given image volume. The second network localizes the actual landmark precisely and accurately in the candidate neighborhood. Both networks take advantage of multi-dimensional gated recurrent units in their main layers, which allow for high model complexity with a comparatively small set of parameters. We localize the medullopontine sulcus in 3D magnetic resonance images of the head and neck. We show that the proposed approach outperforms similar localization techniques both in terms of mean distance in millimeters and voxels w.r.t. manual labelings of the data. With a mean localization error of 1.7 mm, the proposed approach performs on par with neurological experts, as we demonstrate in an interrater comparison.