Machine Learning

We developed a deep learning model for detecting the diagonal sulcus (ds), a tertiary brain sulcus critical for language processing. Using a fine-tuned convolutional autoencoder, our approach achieved state-of-the-art accuracy, outperforming traditional software and other deep learning models. The method leverages self-supervised learning and provides interpretable results aligned with expert annotations, enabling population studies of ds prevalence.

We present a low-cost, automated method for cloud classification using ground-based irradiance measurements and machine learning, achieving 88% accuracy with an XGBoost model. This approach simplifies cloud monitoring, offering an efficient solution for weather services worldwide.

This paper presents a computational tool for guiding intracranial aneurysm treatment by aligning preoperative simulation data with real-time X-Ray imaging, enabling more accurate flow diverter deployment. The method was validated on patient data, showing its potential to improve decision-making during procedures by visualizing the device’s current and planned positions.

Incorporating weak anomaly labels into standard segmentation models improves lesion segmentation results without requiring extra manual labels, enhancing the potential of anomaly guided segmentation for retinal optical coherence tomography scans.

The paper evaluates domain generalization strategies for optic disc and cup segmentation in fundus images, highlighting issues with existing methods when applied to uncropped images, and proposes a semi-supervised learning approach based on the Noisy Student framework to improve performance across diverse datasets.

This paper introduces PALM, an open fundus imaging dataset for pathological myopia recognition, featuring 1200 images with associated labels for pathologic myopia category and manual annotations of optic disc, fovea position, and lesions like patchy retinal atrophy and retinal detachment, aiding in automated diagnostic tools development.

The GAMMA Challenge addresses the need for multi-modality glaucoma grading, providing a dataset with both 2D fundus images and 3D OCT volumes, inviting algorithm development and evaluation, leading to practical insights for clinical diagnosis.

The paper presents a novel method for learning normal asymmetry patterns in brain structures. It accurately characterizes normal asymmetries and detects pathological alterations without relying on diseased cases for training. The approach shows promise in improving the identification of neurodegenerative conditions..

NORHA is a novel index for quantifying hippocampal asymmetries in neurodegenerative conditions. It shows promise as a biomarker for detecting unilateral abnormalities, such as hippocampal sclerosis, and correlates positively with the functional cognitive test CDR-SB, indicating its potential in dementia diagnosis.

We developed a new set of ICL sizing formulae based on LASSO regression, OCT measurements and ocular biometry data.