A5041248487- AI in cancer detection
- Radiomics and Machine Learning in Medical Imaging
- Explainable Artificial Intelligence (XAI)
- Cell Image Analysis Techniques
- Brain Tumor Detection and Classification
- COVID-19 Impact on Reproduction
- Multiple Sclerosis Research Studies
- Machine Learning and Data Classification
- Medical Image Segmentation Techniques
- Digital Imaging for Blood Diseases
- EEG and Brain-Computer Interfaces
- Molecular Biology Techniques and Applications
- Distributed and Parallel Computing Systems
- Cloud Computing and Resource Management
- Gene expression and cancer classification
- Muscle activation and electromyography studies
- Maritime Transport Emissions and Efficiency
- Marine and Coastal Research
- Technology and Data Analysis
- Generative Adversarial Networks and Image Synthesis
- Retinopathy of Prematurity Studies
- Pneumothorax, Barotrauma, Emphysema
- Respiratory Support and Mechanisms
- Hand Gesture Recognition Systems
- Photoacoustic and Ultrasonic Imaging
Icahn School of Medicine at Mount Sinai
2024-2025
IBM Research - Zurich
2022-2025
HES-SO University of Applied Sciences and Arts Western Switzerland
2017-2024
HES-SO Valais-Wallis
2021-2023
University of Geneva
2019-2023
Sapienza University of Rome
2017-2023
Mount Sinai Health System
2022
United States Geological Survey
1986
Since its emergence in the 1960s, Artificial Intelligence (AI) has grown to conquer many technology products and their fields of application. Machine learning, as a major part current AI solutions, can learn from data through experience reach high performance on various tasks. This growing success algorithms led need for interpretability understand opaque models such deep neural networks. Various requirements have been raised different domains, together with numerous tools debug, justify...
Deep Learning (DL) can predict biomarkers from cancer histopathology. Several clinically approved applications use this technology. Most approaches, however, categorical labels, whereas are often continuous measurements. We hypothesize that regression-based DL outperforms classification-based DL. Therefore, we develop and evaluate a self-supervised attention-based weakly supervised regression method predicts directly 11,671 images of patients across nine types. test our for multiple...
Associations between antenatal SARS-CoV-2 infection and pregnancy outcomes have been conflicting the role of immune system is currently unclear. This prospective cohort study investigated interaction infection, changes in cytokine HS-CRP levels, birthweight gestational age at birth. 2,352 pregnant participants from New York City (2020-2022) were included. Plasma levels interleukin (IL)-1β, IL-6, IL-17A high-sensitivity C-reactive protein (HS-CRP) quantified blood specimens obtained across...
The development of automatic segmentation techniques for medical imaging tasks requires assessment metrics to fairly judge and rank such approaches on benchmarks. Dice Similarity Coefficient (DSC) is a popular choice comparing the agreement between predicted against ground-truth mask. However, DSC metric has been shown be biased occurrence rate positive class in ground-truth, hence should considered combination with other metrics. This work describes detailed analysis recently proposed...
Abstract Background Recent meta-analyses showed that around 30% of women suffered postpartum depression during the COVID-19 pandemic, a two-fold increase compared to before pandemic (1,2,3). The immune system plays key role in responding infections, and is potential mechanism development psychiatric disorders. placenta driver changes pregnancy, modulating maternal response perinatal infections (4). Aims & Objectives To investigate: 1) whether SARS-CoV2 infection pregnancy associated with...
Abstract The application of machine learning methods to biomedical applications has seen many successes. However, working with transcriptomic data on supervised tasks is challenging due its high dimensionality, low patient numbers and class imbalances. Machine models tend overfit these do not generalise well out-of-distribution samples. Data augmentation strategies help alleviate this by introducing synthetic points acting as regularisers. existing approaches are either computationally...
Distributional shift, or the mismatch between training and deployment data, is a significant obstacle to usage of machine learning in high-stakes industrial applications, such as autonomous driving medicine. This creates need be able assess how robustly ML models generalize well quality their uncertainty estimates. Standard baseline datasets do not allow these properties assessed, training, validation test data are often identically distributed. Recently, range dedicated benchmarks have...
Computer-aided diagnosis tools for Retinopathy of Prematurity (ROP) base their decisions on handcrafted retinal features that highly correlate with expert diagnoses, such as arterial and venous curvature, tortuosity dilation. Deep learning leads to performance comparable those physicians, albeit not ensuring the same clinical factors are learned in deep representations. In this paper, we investigate relationship between context ROP diagnosis. Average statistics each input image were...
The diffused practice of pre-training Convolutional Neural Networks (CNNs) on large natural image datasets such as ImageNet causes the automatic learning invariance to object scale variations. This, however, can be detrimental in medical imaging, where pixel spacing has a known physical correspondence and size is crucial diagnosis, for example, lesions, tumors or cell nuclei. In this paper, we use deep interpretability identify at what intermediate layers learned. We train evaluate different...
This paper focuses on the uncertainty estimation for white matter lesions (WML) segmentation in magnetic resonance imaging (MRI). On one side, voxel-scale errors cause erroneous delineation of lesions; other lesion-scale detection lead to wrong lesion counts. Both these factors are clinically relevant assessment multiple sclerosis patients. work aims compare ability different voxel- and measures capture related detection, respectively. Our main contributions (i) proposing new that do not...
During the past 60 years scientific research proposed many techniques to control robotic hand prostheses with surface electromyography (sEMG). Few of them have been implemented in commercial systems also due limited robustness that may be improved multimodal data. This paper presents first acquisition setup, protocol and dataset including sEMG, eye tracking computer vision study control. A data analysis on healthy controls gives a idea capabilities constraints procedure will applied amputees...
The representational differences between generalizing networks and intentionally flawed models can be insightful on the dynamics of network training. Do memorizing networks, e.g. that learn random label correspondences, focus specific patterns in data to memorize labels? Are features learned by a affected randomization model parameters? In high-risk applications such as medical, legal or financial domains, highlighting help generalization may even more important than performance itself. this...
Free accessAbstractFirst published online September 30, 2023MSMilan 2023 – ePosterVolume 29, Issue 3_supplhttps://doi.org/10.1177/13524585231196195
Disturbances in T-cells, specifically the Th17/Treg balance, have been implicated adverse pregnancy outcomes. We investigated these two T-cell populations following pre-pregnancy and SARS-CoV-2 infection COVID-19 vaccination 351 participants from a cohort New York City (Generation C; 2020-2022). status was determined via laboratory or medical diagnosis survey electronic records data. Peripheral blood mononuclear cells (PBMCs) were collected at routine prenatal visits throughout gestation...
Mechanistic interpretability aims to understand how models store representations by breaking down neural networks into interpretable units. However, the occurrence of polysemantic neurons, or neurons that respond multiple unrelated features, makes interpreting individual challenging. This has led search for meaningful vectors, known as concept in activation space instead neurons. The main contribution this paper is a method disentangle vectors encapsulating distinct features. Our can...
The increasingly intensive collection of digitalized images tumor tissue over the last decade made histopathology a demanding application in terms computational and storage resources. With containing billions pixels, need for optimizing adapting to large-scale data analysis is compelling. This paper presents modular pipeline with three independent layers detection tumoros regions digital specimens breast lymph nodes deep learning models. Our can be deployed either on local machines or...
Due to energy limitation and high operational costs, it is likely that exascale computing will not be achieved by one or two datacentres but require many more. A simple calculation, which aggregates the computation power of 2017 Top500 supercomputers, can only reach 418 petaflops. Companies like Rescale, claims 1.4 exaflops peak power, describes its infrastructure as composed 8 million servers spread across 30 datacentres. Any proposed solution address challenges has take into consideration...