RESEARCH

What are you interested in?

All topics
AI and Society
Causal Inference
Neuroscience
NLP
Vision
Urban Planning and Smart Cities
Neuroscience and Psychology
AI and Education
Health-Biomedicine
Fundamentals of AI and DS
Environment
Economy and Finance
Digital Humanities
AI, Ethics & Law
AI in the Physical Sciences
AI for social good
Magnetic Sweet AI: Metabolic Brain Cancer Imaging using Deep MRI of a Sugar-Based Contrast Agent

Research

Aug 3rd, 2023
Magnetic Sweet AI: Metabolic Brain Cancer Imaging using Deep MRI of a Sugar-

Researchers: Dr. Or Perlman (Biomedical Engineering) and Prof. Gil Navon (Chemistry)

  • Health-Biomedicine

Despite extensive research efforts, brain tumors remain a leading cause of cancer-related death, with only one-third of individuals surviving more than 5 years after diagnosis. Early detection constitutes a decisive factor in disease prognosis. While various medical imaging modalities can provide an anatomical view of the brain, they only capture morphological tissue changes that occur relatively late, when the tumor is already well-defined and mature.

 

In contrast, alterations in metabolic properties, such as increased glucose consumption, are manifested as early as cell reprogramming and constitute a unique tumor signature. As currently employed in-vivo metabolic imaging techniques require the use of ionizing radiation and have limited spatial resolution, there is an urgent need for an alternative accurate, and safe means for early detection and characterization of brain tumors.

Recently we discovered a new non-toxic sugar-based material, which is preferentially accumulated in tumor cells and can be detected using the chemical exchange saturation transfer magnetic resonance imaging (CEST-MRI) technique. However, the complex molecular brain environment generates a variety of confounding magnetic signals stemming from various metabolites, lipids, and peptides, which severely hinder the quantification of metabolic sugar consumption.

 

The main goal of this project is to develop a transformative radiation-free, and rapid AI-based metabolic MR technology for early brain tumor detection. We propose to adopt a previously unconsidered perspective and to represent the underlying physics of molecular MRI as a computational graph, enabling an automatic AI-based optimization of tumor metabolic imaging.

Using Inverse Reinforcement Learning to Understand What is Being Learned in Motor Learning

Research

Aug 3rd, 2023
Using Inverse Reinforcement Learning to Understand What is Being Learned in

Researcher: Prof. Jason Friedman (Health Professions)

  • Health-Biomedicine

Motor learning is an important part of our lives – when we learn a new skill, during childhood as we learn to control our bodies, and in rehabilitation when we re-learn how to perform day-to-day tasks. It is assumed that we perform these tasks in an optimal way, but it is not obvious what it is that we are optimizing.

 

In this research project, we tackle this problem by using computational techniques, specifically inverse reinforcement learning and inverse optimization, to try to understand what people are learning.

 

We explore this problem at two levels - using inverse reinforcement learning paradigms to determine what reward produces human-like movements in a simple point-to-point reaching task, and using inverse optimization, where we study a complex task (playing the piano) and try to understand what experts are optimizing in terms of selecting posture.

 

These findings can benefit our understanding of how and why we perform particular types of movements, and be used as a tool to accelerate motor learning.

 

 

Summary of the inverse optimization procedure in piano playing comparing experts and novices (a) Visualization of the arms and hands from a motion capture system (b) index finger joint angles when pressing a piano key (c) Slopes of the regression lines (loudness in Sone as a function of position).

 

Positive values correspond to performing a crescendo as required. The expert group performed the task with a higher slope - i.e., took better advantage of the dynamic range. (d) The least comfortable postures selected by the experts were more comfortable than the novices (p=0.05), suggesting that comfort is one factor optimized by experts.

Estimating Real-World Step Length from Inertial Measurement Units Using Advanced Machine Learning Techniques

Research

Aug 3rd, 2023
Estimating Real-World Step Length from Inertial Measurement Units Using

Researchers: Dr. Neta Rabin (Industrial Engineering) and Prof. Jeffrey Hausdorff (Medicine)

We have developed an interdisciplinary model based on machine learning to accurately estimate step length. The new model can be integrated into a wearable device that is attached with tape to the lower back and enables continuous monitoring of steps in a patient’s everyday life.

 

Step length is a very sensitive and non-invasive measure for evaluating a wide variety of conditions and diseases, including aging, deterioration as a result of neurological and neurodegenerative diseases, cognitive decline, Alzheimer’s, Parkinson’s, multiple sclerosis, and more.  Today it is common to measure step length using devices found in specialized laboratories and clinics, which are based on cameras and measuring devices like force-sensitive gait mats.

 

While these tests are accurate, they provide only a snapshot view of a person’s walking that likely does not fully reflect real-world, actual functioning. Daily living walking may be influenced by a patient’s level of fatigue, mood, and medications, for example. Continuous, 24/7 monitoring like that enabled by this new model of step length can capture this real-world walking behaviour.

 

To solve the problem, we sought to harness IMU (inertial measurement unit) systems, which are light and relatively cheap sensors that are currently installed in every phone and smartwatch, and measure parameters associated with walking. "Previous studies have examined IMU-based wearable devices to assess step length, but these experiments were only performed on healthy subjects without walking difficulties, were based on a small sample size that did not allow for generalization, and the devices themselves were not comfortable to wear and sometimes several sensors were needed.

 

We sought to develop an efficient and convenient solution that would suit people with walking problems, such as the sick and the elderly, and would allow quantifying and collecting data on step length, throughout the day, in an environment familiar to the patient.

 

We found that the XGBoost model is the most accurate and is 3.5 times more accurate than the most advanced biomechanical model currently used to estimate step length. For a single step, the average error of our model was 6 cm, compared to 21 cm predicted by the conventional model. When we evaluated an average of 10 steps, we arrived at an error of less than 5 cm, a threshold known in the professional literature as 'the minimum difference that has clinical importance,' which allows identifying a significant improvement or decrease in the subject’s condition. In other words, our model is robust and reliable and can be used to analyze sensor data from subjects, some with walking difficulties, who were not included in the original training set.

Mathematical foundations of xAI and their applications to personalized medicine and public health

Research

Aug 3rd, 2023
Mathematical foundations of xAI and their applications to personalized medicine

Researchers: Dr. Uri Obolski (Public Health) and Prof. Ran Gilad Bachrach (Biomedical Engineering)

  • Health-Biomedicine

Our research is centered on Explainable AI (XAI), wherein we aim to establish mathematical foundations for elucidating model predictions. 

 

The overarching goal is to discern the limitations and possibilities associated with explanations generated by AI models. 

 

By focusing on the development of robust methods, our work contributes to the broader objective of enhancing transparency, accountability, and trustworthiness in AI systems. 

 

The specific applications of this research span diverse fields such as medicine, healthcare, law, and ethics. 

 

Example from Our Research:

 

One of the challenges in machine learning, particularly in medical applications, is accurately interpreting feature importance scores, especially when complex dependencies exist in the data. 

 

A key situation that can lead to unexpected outcomes is the presence of a collider variable, also known as an inverted fork.

 

For instance, consider the example depicted in the attached Figure. Smoking (which is unobserved) increases the likelihood of both Cancer and Gum chewing. Additionally, doctors recommend chewing gum to patients with Earache. When building predictive models for Cancer using features like Gum and Earache, but without knowledge of Smoking, misleading conclusions may arise.

 

  • Scenario 1: A model using only Earache as a feature would assign no importance to Earache since it is independent of Cancer.
     
  • Scenario 2: A model using only Gum would show Gum as an important predictor of Cancer due to their correlation, even though this association arises from the unobserved variable (Smoking).
     
  • Scenario 3: In a model using both Earache and Gum, feature importance scores might misleadingly indicate that Earache is predictive of Cancer. This is because conditioning on Gum (the collider) creates a spurious association between Earache and Cancer.

 

These kinds of situations, known in causality literature, are not easily resolved without additional information about the underlying causal mechanisms. 

 

In medical contexts, such erroneous interpretations could lead to incorrect conclusions, such as associating Earache with cancer risk or wrongly attributing gum chewing as a preventive factor for cancer. 

 

Our research explores both the theoretical underpinnings of these phenomena and provides empirical results to help mitigate such risks when applying machine learning models in healthcare

Generalized regression discontinuity design with multiple time periods and comparison groups for causal inference: theory and practice

Research

Aug 3rd, 2023
Generalized regression discontinuity design with multiple time periods and

Researchers: Dr. Daniel Nevo (Mathematical Sciences), Dr. Itay Saporta-Eksten (Economics), and Dr. Analia Schlosser (Economics)

  • Causal Inference
  • Fundamentals of AI and DS

The widely employed regression discontinuity (RD) causal inference method is intended for scenarios where there is a known cutoff of a running variable such that the probability of receiving treatment changes abruptly at this cutoff. Examples of such settings are countless. The key assumption underlying these analyses is that outcomes of the untreated units just left to the cutoff are good representatives of the counterfactual outcomes of the treated units just right to the cutoff had they been untreated.

 

However, this assumption is often implausible when changes other than the intervention of interest occur at the cutoff (for example, other treatments or policies are implemented at the same cutoff). In these scenarios, researchers retreat to ad-hoc analyses that are not justified by any theory and yield results with unclear interpretations. These analyses seek to exploit additional data for which no intervention was applied for all units (regardless of their running variable value). This could be the case when data from multiple time periods and/or multiple comparison groups are available.

 

In this project, we (1) Develop formal theory and statistical methods for studying causal effects with generalized RD designs utilizing multiple time periods or comparison groups data and (2) Implement the proposed methods to real data and provide practical guidance on how to conduct RD analysis with multiple time points and/or multiple comparison groups.

Tel Aviv University makes every effort to respect copyright. If you own copyright to the content contained
here and / or the use of such content is in your opinion infringing Contact the referral system >>