QUANTITATIVE IMAGING
Learning from
imaging phenotypes.
Quantitative image descriptors and structured biomedical data support machine learning across research questions and cohort sizes.
AN INDEPENDENT RESEARCH INITIATIVE
Medical imaging.
Deeper insight.
Connecting medical images, biomedical data and artificial intelligence to answer meaningful research questions.
From quantitative imaging to multimodal AI.
Built around the question. Grounded in the data.
IMAGING INTELLIGENCEFIG. 01
Medical images → AI → Research insightCONCEPTUAL
A MULTIMODAL PERSPECTIVE
MRI / fMRI / MRACTPET / CT / MRIUltrasoundX-rayEEG & signals
01 / RESEARCH
The right method.
For the right question.
Different datasets call for different approaches. We connect quantitative imaging, computer vision and multimodal learning across three research themes.
Learning from
MEDICAL COMPUTER VISION
Learning directly
from images.
Image-based methods for segmentation, classification and representation learning, from 2D acquisitions to 3D volumes.
MULTIMODAL INTELLIGENCE
Connecting images
and context.
Exploring how medical images, structured phenotypes and clinical text can inform shared representations and image-to-text research.
IN FOCUS
Research directions
01Neuroimaging & brain researchRESEARCH THEME
Exploring artificial intelligence for brain imaging, with an interest in understanding complex patterns and supporting reproducible neuroscience research.
02Multimodal image understandingEXPLORATORY DIRECTION
Exploring segmentation, learned representations and fusion across imaging modalities. Study design and model selection are tailored to each dataset and research question.
03Medical vision–language learningEXPLORATORY DIRECTION
Exploring relationships between medical images and clinical text through multimodal representations, image-to-text generation and research-oriented reasoning.
02 / OUR APPROACH
From raw data.
To research you can build on.
A connected workflow, with quality, validation and reproducibility considered at every stage.
01 / PREPARE
A strong study starts
with well-understood data.
Review cohort structure and acquisition quality, define the target task, and establish modality-specific preprocessing.
- Image quality control
- Registration & standardisation
- Cohort & outcome review
02 / REPRESENT
Find the representation
that serves the question.
Build quantitative image descriptors or learn representations directly from images, according to the needs of the study.
- Segmentation & regions of interest
- Quantitative feature extraction
- Learned image representations
03 / MODEL
Choose the model
around the study.
Develop tabular, image-based or multimodal models according to sample size, available labels and the intended research outcome.
- Feature selection within training folds
- Classification & regression
- Multimodal modelling
04 / VALIDATE & INTERPRET
Test the result.
Understand its limits.
Evaluate generalisation with appropriate cross-validation, centre-aware testing and independent data where available.
- Internal & external validation
- Error analysis & feature attribution
- Documented code, figures & methods
Transparent methods. Documented experiments. Reproducible analysis.
THE PRINCIPLE THROUGHOUT03 / COLLABORATE
Your research question.
Our shared next step.
Have imaging data and a question worth exploring? Let’s discuss how computational imaging and AI could help.
Selected academic projects may receive support at no cost, subject to scope, feasibility and available capacity.
hello@leoimagelab.onlineA conversation first.
- 01
Share your question
Tell us the modality, cohort size, research goal and available outcomes.
- 02
Explore the fit
We discuss scientific scope, feasibility, ethics and data-use requirements.
- 03
Shape the study
Agree on a workflow, validation strategy and research outputs together.
Start with a project summary. Please do not email identifiable patient data.
04 / ABOUT THE LAB
Curiosity, with
computational rigour.
Leo Image Lab is an independent research initiative focused on artificial intelligence for medical imaging and multimodal biomedical data.
Our aim is to connect useful methods with meaningful scientific questions — through thoughtful study design, transparent evaluation and collaborative research.
RESEARCH CORRESPONDENCE905 Armstrong St
Ballarat VIC 3350, Australia
Ballarat VIC 3350, Australia