Ian Pan

56 posts

Ian Pan

@ianpanmd

MD, PGY-3 Radiology @BWHRadEdu Kaggle Grandmaster, healthcare + AI Views my own.

Boston, MA Katılım Nisan 2020

58 Takip Edilen1.3K Takipçiler

Sabitlenmiş Tweet

Ian Pan@ianpanmd·21 Ağu

The medals have been awarded, and it's official! I finally became a Kaggle Grandmaster, which I never thought would happen. I have to thank @FelipeKitamura, @alexandrecadrin, and @DrJHoward for carrying me to my first 4 medals.

English

215

Ian Pan@ianpanmd·3 Kas

Also amazing how easy it was to put together this demo! Should be a standard moving forward for radiology AI publications.

English

Ian Pan@ianpanmd·3 Kas

Pediatric bone age model (G&P) now on @huggingface Spaces: huggingface.co/spaces/ianpan/… Trained on the @RSNA 2017 challenge data, MAE 4.45 on the test set. Demo purposes only. I am not responsible for how you use the output of this model. 🤗

English

Ian Pan@ianpanmd·30 Eki

Excited to have finished in the top 10 in this year's @RSNA @TheASNR @The_ASSR C-spine Fracture Detection AI Challenge on @kaggle! Congrats to all of the other winners & competitors, and big thanks to the organizers and annotators. Solution summary here: tinyurl.com/4zayhwnp

English

Ian Pan@ianpanmd·11 Nis

@FelipeKitamura @falsalem76 @drsxr @clindatsci @Radiology_AI @tunguz @abhi1thakur Differences in scanners, acquisition, patient population, disease prevalence, etc make it hard for models to generalize. Hard to address. I'd imagine even human radiologists at 1 hospital would initially underperform at another if all of the above are very different.

English

Felipe Kitamura@FelipeKitamura·10 Nis

@falsalem76 @drsxr @clindatsci @Radiology_AI I wonder what @tunguz, @ianpanmd, and @abhi1thakur think about this scenario.

English

Felipe Kitamura@FelipeKitamura·10 Nis

Training on data from institution A did not perform so well on institution B and vice-versa. Even if both institutions had a diversity of MRI scanner manufacturers. Training on A+B made it generalize to test sets A and B. @clindatsci #Dasa @Radiology_AI pubs.rsna.org/doi/abs/10.114…

Hari Trivedi@HariTrivediMD

What happens when deployed in 10 US cities with wildly different demographics and age distributions? What if scanner firmware or image post processing is updated? How many cases will each site audit annually to track performance? What’s the acceptable miss rate for the model?

English

Ian Pan@ianpanmd·7 Şub

Not sus at all when a reviewer tells you to cite 3 loosely related papers by the same person 🤔

English

Ian Pan retweetledi

Walter Wiggins, MD, PhD@walterfwiggins·27 Kas

Mon 11/29, 4:30pm & Wed 12/1, 9:30am: Oooh, a double-feature! Not one, but 2⃣ chances to have @kaggle winners @FelipeKitamura from @UNIFESP_Rad & grandmaster @ianpanmd from @BWHRadEdu show you how they won the pneumonia detection challenge. #RSNA21

GIF

English

Ian Pan@ianpanmd·15 Tem

@JFPuget @DrLukeOR Good point!

English

JFPuget 🇺🇦🇨🇦🇬🇱@JFPuget·14 Tem

@ianpanmd @DrLukeOR This says more about the ability to calibrate confidence by humans than ability to diagnose IMHO. F1 score with binary predictions would have been more interesting.

English

Ian Pan@ianpanmd·14 Tem

thelancet.com/journals/landi… Radiologists: 0.713 AUC Radiologists+AI: 0.808 AUC! AI alone: 0.957 AUC ...? Should I take back my previous tweet? 😂 @DrLukeOR

Français

Ian Pan@ianpanmd·9 Tem

Just finished 1st week of radiology! 😅 There's no way AI is going to replace us anytime soon lol

English

584

Ian Pan@ianpanmd·6 Ara

@drsxr @FelipeKitamura @kaggle @abhi1thakur @alexandrecadrin That's not unique to Kaggle, as Alex said it applies to ML at large. Also unclear what that has to do with ensembles.

English

Just another south FL dude@drsxr·5 Ara

@FelipeKitamura @kaggle @abhi1thakur @ianpanmd @alexandrecadrin That’s my complaint about @kaggle - it’s curve fitting. Otherwise why does every one have to ensemble everything? (Good science sure but Alex & Ian know what I mean)

English

Felipe Kitamura@FelipeKitamura·5 Ara

In @kaggle competitions, how often do you trust your CV more than the public LB? Feel free to comment. @abhi1thakur @ianpanmd @alexandrecadrin @drsxr

English

Ian Pan@ianpanmd·27 Kas

@AmineKorchiMD If you give me a labeled dataset, I'll give you the model!

English

Dr Amine Korchi@AmineKorchiMD·26 Kas

A software that automatically identify the rib level on any #CT plane would be hepful when reporting a rib lesion such as a fracture ! #AI #radiology

English

Ian Pan@ianpanmd·26 Kas

@ParasLakhaniMD @RSNA @kpa_kathy @kaggle @thoracicrad Thank you!

English

Paras Lakhani@ParasLakhaniMD·26 Kas

Congrats to @ianpanmd for winning the Educational Merit Award for the RSNA-STR PE Kaggle Challenge for the clarity and comprehensiveness of his explanation @RSNA @kpa_kathy @kaggle @thoracicrad. Well done! press.rsna.org/timssnet/media…

English

Ian Pan@ianpanmd·19 Kas

@FelipeKitamura @SafwanHalabi No, my take from this paper is that no matter how robust your CV scheme is you can still have poor model performance on unseen test data

English

Felipe Kitamura@FelipeKitamura·19 Kas

@SafwanHalabi @ianpanmd, does your nested CV approach help mitigate this underspecification issue?

English

Safwan Halabi@SafwanHalabi·19 Kas

More nuances to the "Black Box" of AI models ... lnkd.in/gMEntQK

English

Ian Pan@ianpanmd·1 Kas

Also @unifesp and @FelipeKitamura who've also provided data for the past 2 RSNA challenges. I know Felipe loves to Kaggle but chooses to share data from his institution even if it means he can't compete.

English

Ian Pan@ianpanmd·1 Kas

Most of my success is 2/2 open access medical imaging. I'm super grateful to those who've made data sharing a top priority. In particular, SO to @StanfordAIMI @StanfordRad who've proven time and time again that HIPAA and academic bureaucracy are not insurmountable obstacles.

English

Ian Pan@ianpanmd·30 Eki

If anyone wants to listen to me ramble for 30 minutes 😂

Radiology@radiology_rsna

In the latest episode of the @Radiology_AI podcast, @PaulYiMD interviews @ianpanmd on machine learning, data science and becoming the first physician @Kaggle Grandmaster. bit.ly/RadAiPodcasts #RadAI #RadPodcasts

English

Ian Pan@ianpanmd·29 Eki

Results are official! I'm happy to have won 2nd place in the @RSNA @thoracicrad PE Detection challenge on @kaggle. I've also sneaked into the top 10 for competitions. Read more about my solution here: kaggle.com/c/rsna-str-pul…

English

173

Ian Pan@ianpanmd·28 Eki

99% sensitive, 99% specific model for finding with 10% prevalence has PPV of 91.7% = ~1 FP/12 predicted +. 5% prevalence, PPV 83.9% = ~1 FP/6 predicted +. 5% prev, 95% sens, 95% spec, PPV 50.0% = 1 FP/2 predicted +. Pre-test probability is important.

English

Ian Pan@ianpanmd·23 Eki

One issue with separate models for each minority group is that now you have to assign patients to groups and use the corresponding model. A lot of bias and assumptions go into assigning patients into groups.

Curt Langlotz@curtlanglotz

"Prevalent schemes of multiethnic machine learning are prone to generating significant model performance disparities between ethnic groups. These disparities are caused by data inequality and data distribution discrepancies between ethnic groups." nature.com/articles/s4146…

English

Ian Pan@ianpanmd·20 Eki

Agree 100%. I hate saliency maps.

Neil Tenenholtz@ntenenz

@quantrad Trying to interpret saliency maps is asking for trouble. The example I like to use is detecting a brain tumor via midline shift (or vice versa). My go-to recommendation for classification interpretability is to actually train a segmentation model and use that to classify.

English

Keşfet

@huggingface @RSNA @TheASNR @The_ASSR @kaggle @FelipeKitamura @falsalem76 @drsxr