🎉 Honored to receive the Best Research Paper Award at
#SIIM23
! 🏆
Huge thanks to the amazing team, esp.
@PRouzrokh
who deserves equal recognition for his invaluable contributions. 🙌
Grateful to
@Slowvak
&
@CodyWyles
for their exceptional support.
@SIIM_Tweets
I am thrilled to share that I will start my radiology residency at the Yale University! This is an incredible moment for me & the start of a very exciting journey! Immense gratitude to
@YaleRadiology
for this tremendous opportunity and their trust in me!
#Match2024
#futureradres
Thrilled and honored to have been selected as a Trainee Editorial Board member for
@Radiology_AI
#TEB
.
Looking forward to learning from experienced mentors of the field of radiology and Al over the next two years! Thank you
@judywawira
and
@cekahn
for this valuable opportunity.
🩻 Super excited to announce that our abstract has been accepted for the
#RSNA23
cutting-edge research session!
This work is near and dear to me, and I’m so happy I can share it at this great podium.
Grateful for all the colleagues and mentors who have made this possible 🙏🏻
Thrilled to share our preprint on use of
#SyntheticData
in medical imaging research!
📝
💬 Here is a custom GPT that will answer all your questions about the paper:
Some key takeaways in a 🧵 /1
Last year’s
#RSNA22
was an unforgettable experience 🌟, the largest and most thrilling event I’ve ever attended!
Eagerly anticipating
#RSNA23
, ready for an incredible week of learning 📚, reuniting with friends and mentors 👥, and making many new connections! 🤝
I had a blast at the
#AIBiasDatathon
! It was so much fun to work with this massive data, see perspectives from all around the world and catchup with friends.
A big shout-out to
@judywawira
@HariTrivediMD
,
@angiowoman
and the whole team for making this possible.
"=" > "÷"
🎄Check out our latest work on
#GenerativeAI
published recently at
@JArthroplasty
. Here we use
#GANs
to generate synthetic radiographs and examine their usefulness in inter-institutional data sharing.
Check out the full text:
A 🧵 on what is happening in
#Iran
and what we all can do;
You all might have seen the global # trend,
#Mahsa_Amini
(in farsi #مهسا_امینی). In this thread, I will try to explain the situation to my non-Iranian network and finally ask for their help.
1/n
Still buzzing with excitement! 🎉
I'll be doing 3️⃣ oral presentations, sharing 1️⃣ digital poster, and 1️⃣ educational exhibit at
#RSNA23
! Thank you
@RSNA
! And thanks to all the mentors and colleagues 🙏🏻
See you in chilly November, Chicago! 🏙️
🌟 Exciting news!
@Radiology_AI
Living Resource Index: a go-to resource for
#publicly_available
datasets in radiology.
🩻 Next time you’re training a model or validating an algorithm, save time and discover a wealth of diverse datasets through this resource.
🚀Happy to introduce `nb2py`, my side project from this past week that I've wanted to tackle for a looong time! 🛠️
It's here to bridge the gap between Jupyter Notebooks and training-ready Python scripts.
#DeepLearning
#DataScience
🥁 Thrilled to share our work titled `RadRotator`, a
#Diffusion
model for rotating radiographs in 3D space with maximal user control! 🩻
Last year we released `Mediffusion`, the backbone of our
#GenerativeAI
work!
Check it out to train your own👇🏻
Excited to share the technical report for RadRotator, our latest
#GenerativeAI
tool, which enables the rotation of radiographs in 3D space💫
✅Technical Report:
🔗
✅Website:
🔗
✅Online Demo:
🔗
(1/3)
It was an amazing start to
#DDW2024
!
Exited to have presented the result of our study evaluating the prognostic value of morphological features for prediction of outcome in patients with
#colorectal
cancer using
#DeepLearning
.
I have been selected as a new member of
@SIIM_Tweets
#ML
#education
subcommittee. Look forward to working with this awesome team!
Thanks to Drs
@FelipeKitamura
& Timothy Kline for this opportunity.
🚀
@TheoDapamede
and I are thrilled to announce
#RadPrompter
- a Python package for simplified, reproducible, and composable
#LLM
prompting, providing a flexible framework for crafting complex prompts from reusable components.
1/7
#Embeddings
are a way to compress image information, but are they on par with original images? Do they encode protected information?
This was our teams project! Stay tuned for more amazing things from these folks!
Good reading source:
#AIBiasDatathon
Congratulations Team 4/11 1st place
#AIBiasDatathon
! Amazing teamwork Alexander Sheih, Cheng Ding, Celestino Obua, Jasmit Shah, Renata Proa, Swapnil Mishra, Dana Moukheiber, Martin Galiwango, Bardia Khosravi, Saptarshi Purkayastha, Andrew Sellergren, Lama Moukheiber, Ran Xiao
⚕️Medicine is a field that is inherently
#multimodal
.
🩻 Over the years, AI systems have achieved high levels of performance on specific tasks using a single modality (mostly images).
🌟 Join us in the next
#RadAIchat
to talk about the next generation of
#MultimodalAI
models.
🪩 Check out our latest work on
#DiffusionModels
and their ability to do a "virtual" surgery, led by the one and only
@PRouzrokh
!
My favorite part? The evaluation procedure and quality assessment by surgeons!
Read more in the full text 👇🏻
1) 🔥Excited to share our recent publication at
@JArthroplasty
, which introduces THA-NET, our new
#DiffusionModel
at
@Mayo_OSAIL
&
@MayoAILab
to mimic high-quality total hip arthroplasty on plain radiographs.
Live demo:
Details and link to article: 🧵👇
After the first FDA-approved total hip arthroplasty (THA) was performed at Mayo Clinic in 1969, the clinic began a total joint arthroplasty registry to follow all patients’ outcomes throughout their lifetimes.
@Slowvak
@CodyWylesMD
@Khosravi_Bardia
Hot off the press 🔥
In this study, we used a hybrid CNN-transformer architecture to predict risk of implant dislocation after total hip arthroplasty (THA) 🦴
Thanks to my mentors,
@CodyWylesMD
&
@Slowvak
, for their support and guidance.
@MayoAILab
@Mayo_OSAIL
@MayoRadiology
We at
@MayoAILab
prepared a series of papers/guidelines on mitigating bias in
#Radiology
#Machine_Learning
, covering three main areas of focus, data handling, model development, and performance evaluation!
You can find them in
@Radiology_AI
:
Last couple of weeks I was focusing on learning more about the
#NLP
side of AI and it was so much more fun than I expected!
Today I was fortunate enough to summarize what I learned and present it to my brilliant colleagues
@MayoAILab
.
Thank you
@Slowvak
for the opportunity.
Hot off the press 🔥
Check out our recent publication in
@radiology_rsna
! We review some concepts of
#uncertainty
#quantification
and explain why it is important in medical imaging
#ML
research!
Stay tuned for more...
We are humbled by this enthusiasm at
@NVIDIAGTC
!
Unfortunately, our DLIs have a maximum capacity of 80 and many were not able to attend the live workshop. Recordings will be available soon.
See you tomorrow, 3pm CST at the
#Uncertainty_Quantification
learning lab.
🎉 Congratulations,
@ShahriarFaghani
, on winning the title of the most valuable JDI reviewer! 🏆 Your hard work, dedication, and expertise have earned you this honor. Well-deserved recognition for your outstanding contributions!
@SIIM_Tweets
@MayoRadiology
#SIIM23
Meanwhile in Iran, more than 80 men and women have been killed in
#IranProtests
. There is
#internetshutdown
and this number is just an estimate based on estate media; who knows what the real number is😔. Be our voice.
#Mahsa_Amini
#مهسا_امینی
We are excited to share our latest
@GoogleHealth
work to build lightweight multimodal generative AI models for radiology. By "grafting" language-aligned vision encoders onto a fixed LLM, we are able to solve previously challenging tasks using an approach that we call ELIXR...🧵
Great
#SIIM23
keynote by
@oziadias
on the importance of
#unstructured
data hidden in medical images, also how
#OpenScience
can help with validation.
"Let’s explore the potential of emerging technologies to revolutionize our world in ways we have yet to imagine."
@SIIM_Tweets
Fantastic session about career development as a future radiologist with two talks about flexible work arrangements, by
@ParisaMazaheri
, and strategic alignment, by
@AGrayev
.
#ASNR23
This is an AMAZING paper!
Basically showing how *mistakes* in data handling and model development can make an ML model with actual AUC of 0.55 have and apparent AUC of 1.0.
The authors do lots of experiments to pinpoint some of the most important issues, fantastic!👌🏻
🥁🥁🥁
Get ready for a captivating journey into the world of AI and Bias!
@hiti_lab
is hosting an event that’s all about understanding and addressing Bias in AI. With a stellar lineup and thought-provoking insights, it’s bound to be a must-attend! Don't miss it!
🥁🥁🥁
One small fun fact that you might not know about
#ChatGPT
: It is multilingual!
Here I'm asking it (in Persian) to write a letter to ask for a leave of absence in English, and it does one heck of a job!
We just added a new feature that gives you the ability to preview a wide range of plot types for any pandas dataframe. Choose one you like and we’ll auto-insert the code into your notebook. Read more at the Colab blog:
Never thought I would stop using Chrome, let alone replacing it by
#Edge
!🤷🏻
Note that this is Edge dev which is incomparable to the public version, it is insanely robust, plus it has the famous Bing+AI.
@Radiology_AI
T4.
@MayoAILab
has released an online free “book” called MIDeL (Medicsl Imaging & Deep Learning). There are 13 beginner chapters available so far.
#RadAIchat
Amazing talk by
@GregZ_MD
about applications of
#synthetic
MRI in neuroradiology.
Interestingly, synthetic images are found to be more favourable compared to the original ones!
#ASNR23
Starting this December, I will be taking on the role of Associate Editor for Radiology: Artificial Intelligence
@Radiology_AI
.
I sincerely thank our Editor, Dr. Charles Kahn, for trusting me with this role. I'm excited for this new chapter to begin!
Check out this
#LLM
#prompting
guide by
@AliTejaniMD
with amazing examples:
Specifically check the Chain-of-Thought (
#CoT
) prompting, I've seen huge improvements with it. Definitely give it a try if you haven't already!
@Radiology_AI
Google struck back!
CT report generation (3D) leveraging the long context length (>1M) of Med-Gemini (🤯) among other 2D tasks.
Should be a very interesting read…
Excited to share latest ✨Med-Gemini✨ additions - our new research unlocks possibilities in medical data analysis with 3 new models built upon Gemini 1.5 that can handle 2D medical images, and for the first time genomic risk score & 3D radiology scans.
🤩 So excited and humbled that
@Pair_Medicine
picked up our work recently published in
@Radiology_AI
.
They are the best at simplifying medical
#AI
literature and explaining it to healthcare professionals.
It is usually overlooked that
#ChatGPT
is a wrapper around the original model (GTP-3.5 or 4), meaning there is a *system* prompt injected by OpenAI and custom inference configurations (e.g. temperature), which makes it undeterministic!
At least use the
#API
in research projects.
🤨 "54% accuracy" ➡️ "
#ChatGPT
shows potential as a valuable diagnostic tool in
#radiology
"
🤔 Because of a median legitimacy score of 5/5?
😞 Another openbook test paper & no prompts or examples
#GenAI
#GenerativeAI
Denoising Diffusion Probabilistic Models (
#DDPMs
) are a recently introduced (relative to GANs) group of generative models.
Check out our discussion with
@PRouzrokh
about what these models are, and some of their potentials in medical image research.
You have heard of
#DALLE2
, the model that creates realistic photos based on text prompts! This is an example of "denoising diffusion probabilistic models".
#DDPMs
Check out the video by
@PRouzrokh
&
@khosravi_bardia
to learn more about them and their applications in medicine!
Very engaging talk by
@judywawira
on AI
#fairness
&
#bias
.
To sum up models are good at picking up small differences between different groups, and this will harm the minorities, so start mitigating the bias from the moment you start gathering your data.
@SIIM_Tweets
#Virtual
Don't forget the May 5th deadline for applying for the
@Radiology_AI
#TEB
! 📅✨
It has been a transformative experience for me over the past year, and I've found great mentors and fantastic friends. Feel free to DM me if you have any questions about it! 💬
📒 In this hands-on lab you will learn how GANs work, how to code a simple GAN, and how GANs are being used in radiology research.
❇️ Hopefully by the end of this session you will understand what this image represents and how to create one yourself:
Join expert speakers for our
#AI
Webinar in 1 hour & gain an informed perspective on the potential impact of generative
#AI
on patient care in radiology & medical imaging!
Aug 16 | 1:00 PM ET |
#SIIM
Members Earn 1 HR SIIM IIP CE
Sign Up ➡️
Come and join our hands on deep learning workshop on
#UncertaintyQuantification
.
There are lots of amazing sessions in this
@NVIDIAGTC
, keep 👀 on them.
GTC is free and virtual, you just need to sign up here:
You can try your hand at
#GPT4
and work together with others to build a prompt that solves some really tricky problems.
Check out the link to learn more about the activities that will happen at the
#AI_Playground
.
We've released the Rᴀᴅ-DINO model weights!!
Benchmark it, encode some datasets, show us some UMAPs, plug it into your classifiers, LLMs, MLMs, SLMs...
We're excited to discover what the community will create on top of Rᴀᴅ-DINO.
🤗
@MSFTResearch
👨⚕️ Ali has a knack for business and has explored various paths during and after med school: 📚🔬🏥📈
💡 From research to practicing medicine and even starting his own business, he offers a unique perspective. Vote now to hear his point of view at
#RSNA23
#Fast5
! 🤩
So proud to be a part of this fantastic project. Humbled by this recognition🥇
In this educational review, we cover the pathophysiology of acute bowel ischemia and its radiologic findings in different stages:
@ESR_Journals
#Insights_Into_Imaging
🏅 A final congratulations to the
#InsightsImaging
ESGAR Award winners for their article “Many faces of acute bowel ischemia: overview of radiologic staging”!
A wonderful accomplishment!
🎉 (A.H. Davarpanah, A.G. Khameneh, B. Khosravi, A. Mir, H. Saffar & A.R. Radmard)
I highly recommend curating a catalog of the imaging studies and creating a clean
#imaging_registry
. This has saved us a lot of time at
@Mayo_OSAIL
and lowered our
#data_curation
time from couple of weeks to less than an hour!
Thanks to
@PRouzrokh
for leading this fantastic work.
External validation is the key to ensure an AI model actually "works". Join
@ShahriarFaghani
for his presentation of his work on Barret's esophagus detection!
#SIIM23
@SIIM_Tweets
If you're attending
#SIIM23
, don't miss
@ShahriarFaghani
oral presentation on AI in Enterprise Research Abstracts Sessions. Join him at Room 9, Level 3 from 10:00 to 10:45 AM.
@SIIM_Tweets
A roadmap to implement AI in clinical practice:
- Who decides what should be implemented?
- How to assess potential applications of implementation?
- How should a tool be implemented?
- How should an implemented tool be monitored?
1/ Struggling to get 3D models to perform well on 3D medical images like chest CTs?
Our latest research at Harvard+Stanford explores the potential of video-pretraining to unlock 3D model performance and outperform 2D baselines.
#MIDL2023
Check it out 🧵
#RSNA22
gave
@MayoRadiology
alumni and staff from around the country an opportunity to gather, connect and reconnect, and celebrate. Thank you to all who were able to attend!
#MayoAtRSNA22
This is a serious issue with AI generated content.
Other than this Bloomberg post, here is an interesting paper focusing on
#StableDiffusion
inherent biases:
#Bias
#GenerativeAI
🚨Generative AI has a serious problem with bias🚨
Over months of reporting,
@dinabass
and I looked at thousands of images from
@StableDiffusion
and found that text-to-image AI takes gender and racial stereotypes to extremes worse than in the real world.
🧵 1/13
Give this a try (with some medical images), it is 🤯!
I am thinking for a first pass, rough annotation draft this can be used very effectively. Based on the paper, each image can be segmented in 50 milliseconds (once embedded).
Today we're releasing the Segment Anything Model (SAM) — a step toward the first foundation model for image segmentation.
SAM is capable of one-click segmentation of any object from any photo or video + zero-shot transfer to other segmentation tasks ➡️