Ricardo Mokhtari

About Me

Hi, I’m Ricardo. I am currently an AI Research Scientist at AstraZeneca in Cambridge, UK. At AstraZeneca I have focused on delivering disease insights using state of the art Machine Learning approaches to genomic and medical imaging data, with a focus on model explainability. My most recent work includes developing interpretable Computer Vision models for predicting disease relevant features from images of tissues using self-supervised learning and vision transformers (currently accepted for publication at MIDL).

Prior to joining AstraZeneca, I graduated from Imperial College London with a 1st Class (Dean’s List) MEng degree in Molecular Bioengineering. During my time at Imperial I was a part of the Biological Control Systems Lab, led by Dr. Reiko Tanaka, where I researched the use of state of the art generative models (StyleGAN, Pix2Pix, VAEs) as a data augmentation technique for improving the predictive performance and adversarial robustness of deep Computer Vision models. I was also a part of the Advanced Data Science Team, where I worked on an industrial research project with Refinitiv on autonomous web crawling using Reinforcement Learning.

In my free time I hugely enjoy Sci Fi (books, movies and video games), books about Biology and Theoretical Physics (right now reading Nick Lane and Jim Al-Khalili), weight lifting and swimming.

Experience

AI Research Scientist | AstraZeneca

Sept. 2021 - Present

Project 1: Applying self-supervised learning to medical imaging [Accepted at MIDL 2023] Code

Developed and applied state of the art open-source computer vision methods to 1000s of gigapixel images to inform translational medicine teams within AstraZeneca (work showcased to EVP level and presented research to Global Product Team) – pushed AUC from 0.6 to 0.87
Proactively communicated research findings to unfamiliar and non-technical audiences
Collaborated extensively with interdisciplinary teams including AI scientists, image analysts and pathologists, using Agile framework

Project 2: Continual active learning platform for medical imaging

Developed robust infrastructure for deploying, continuously monitoring and improving computer vision models for 3 internal stakeholders
Deployed models on server backend using MONAI, built a GUI using Dash to serve users
Proactively gathered user requirements, refined solution using Agile methodology

Project 3: Using graph machine learning to discover new cancer biomarkers

Leveraged multi-modal genomic dataset to identify novel cancer subtypes and associated biomarkers using graph machine learning and presented to translational teams to inform 2023 oncology R&D strategy
Explored and demonstrated that graph approaches are powerful for multi-modal cancer datasets - pushed AUC from 0.83 to 0.89

I have proactively championed a data-driven culture at AstraZeneca by:

Organising the first AZ Hack (organising team of 3) – a global-scale Data Science hackathon attended by 173 participants across 15 countries
Writing (from scratch) and delivering a 3-hour computer vision workshop to 50 AZ employees
Teaching an 8-week Python course to a class of 60 AZ employees

Data Scientist | Imperial College London + Refinitiv | `Code`

Nov. 2020 – May 2021

Industrial Data Science research project with Refinitiv, a large financial services company
Selected to be part of Imperial’s Advanced Data Science Team - developed data-driven methods for autonomous web crawling using reinforcement learning
Co-developed intelligent web crawling strategy from scratch, co-wrote technical reports and delivered presentations to managing directors at Refinitiv

Research Assistant | Imperial College London | `Paper`

Oct. 2019 – Jun. 2021

Explored and evaluated the utility of using generative models as a data augmentation technique for boosting the performance and robustness of computer vision models
Devised a simple framework for quantitatively evaluating model robustness, showed that generative models are a successful approach
Developed and applied SOTA models (StyleGAN, Pix2Pix, VAE)

Marketing Intern | Ordetto

May. 2019 – Jun. 2019

Independently carried out market research, cold called 200+ businesses and secured pitch meetings with managers at luxury hotels and multi national corporations
Designed and wrote consumer survey, processed and delivered data to management

Education

Imperial College London | MEng Molecular Bioengineering

Oct. 2017 – Jun. 2021

Grade: First Class Hons. (74.93%)
Dean’s List 2021 – Prize for scoring in top 10% of students

Charterhouse School

2012 – 2017

A-Level: A* A* A A
GCSEs: 11 A*s
Academic scholarship worth £1500/year

Publications

R Mokhtari et al., Interpretable histopathology-based prediction of disease relevant features in Inflammatory Bowel Disease biopsies using weakly-supervised deep learning. [Accepted for MIDL 2023]
R Mokhtari et al., Predicting disease relevant features in Crohn’s Disease and Ulcerative Colitis from Haematoxylin & Eosin stained whole slide images using self-supervised deep learning, Journal of Crohn’s and Colitis 2023, https://doi.org/10.1093/ecco-jcc/jjac190.0407 (Impact factor >10)
Attar, R., Hurault, G., Wang, Z., Mokhtari, R., Pan, K., Olabi, B., Earp, E., Steele, L., Williams, H. and Tanaka, R.J., 2022. Reliable detection of eczema areas for fully automated assessment of eczema severity from digital camera images. medRxiv.
Hurault, G., Pan, K., Mokhtari, R., Olabi, B., Earp, E., Steele, L., Williams, H.C. and Tanaka, R.J., 2022. Detecting eczema areas in digital images: an impossible task? JID Innovations, p.100133.