Live and dead cells show distinct morphological signatures in brightfield images, and a model trained on stained cells generalizes to unstained ones. Building on this, we present Viabilense, a deep learning pipeline for label-free cell counting and viability prediction.

We pose fiber bundle image reconstruction as an inverse problem, where the measurement is a blurred, fiber-masked version of the underlying scene. The high-resolution image is then recovered by optimizing a coordinate-based network at test time.

We present an AI-based method for cell viability assessment directly from brightfield images, without staining or fluorescent imaging. We demonstrate its efficiency in CHO cells under a wide range of viability conditions.

We presented a dimensionality-reduction framework for multimodal microscopy that integrates information from different imaging modalities into a shared low-dimensional manifold and demonstrated its effectiveness for live/dead cell phenotyping.

We introduce an unsupervised method with implicit neural representation that reconstructs high-resolution fiber bundle images from misaligned bursts without calibration or paired data.

We developed an open-source ECG-based machine learning pipeline to distinguish hypertrophic from dilated cardiomyopathy subtypes using standard and vectorcardiogram features.

We use nonlinear dimensionality reduction for unsupervised, personalized heartbeat phenotyping from ECGs and arrhythmia detection.

We propose a snapshot hyperspectral imaging method using randomized aperture codes and implicit neural representations.

We reviewed recent advances in self-powered wearable devices, covering energy harvesting techniques, power management, and their application in long-term health and wellness monitoring.

We developed an ultra-low-power, lightweight ECG patch for long-term monitoring, paired with an AI-assisted cloud platform for denoising and arrhythmia detection.