Richard RichardObi

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Hi, I'm Richard 👋🏾

_{Teaching models to create and segment medical scans, and worrying about whether the results can be trusted.}

I build generative models for medical imaging, synthesising MRI, mammograms, and CT, sometimes to sidestep a contrast injection, sometimes just because real data is scarce and locked away. And because fake medical images are only useful if you can trust them, I like to spend a lot of time on how to measure that. 🔬

These days I'm a postdoc working on AI for radiation oncology at TUM in Munich, trying to get vision–language models to help plan cancer treatment. Before that I did my PhD on generative models for breast imaging in sunny Barcelona 🌇, and before that I spent a few years at IBM building AI systems (and accidentally collecting a few patents like this one along the way).

🧰 Things you might find useful

If you're building in medical imaging, some of these might be useful, but if not, reach out to let me know what we should be building together:

🩻 medigan — a model zoo of pretrained generative models for medical iamging. pip install medigan, pick a model, get synthetic mammograms, MRI, x-rays, or endoscopy images. Built so labs can share models instead of locked-away data. → docs here

📏 frd-score — a way to measure whether your synthetic medical images are any good, using radiomic features instead of metrics borrowed from natural image photos. → project page

💉 ccnet & SimulatingDCE — teaching models how contrast flows through breast tissue over time, so we might one day need fewer injections by turning non-contrast MRI into contrast-enhanced MRI → paper

🔒 mammo_dp — Making differential privacy useful with synthetic data - which actually helps a lot when training differentially private cancer classifiers. → paper

🌱 A few things I care about

Making good tools open, so the next person doesn't start from zero
Helping students and newcomers get their first medical-imaging models running
Fair AI that works for everyone, not just the patients who look like the training set — I helped to create a guide to be aware of such potential biases called FUTURE-AI.
Bringing people together around shared problems — I organise the MAMA-SYNTH challenge at MICCAI to reduce the reliance on contrast agents 🏁

💬 Come say hi

If any of this is useful to you, or you're stuck on something, open an issue or reach out. I'm always happy to help you get results.

_{🏓 Also: if you're ever at a conference and want to lose at table tennis, I'm your guy.}

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!

Richard RichardObi

Hi, I'm Richard 👋🏾

🧰 Things you might find useful

🌱 A few things I care about

💬 Come say hi

Pinned Loading

Uh oh!