6/ The big win for automated 🤖matching is in time savings. Automated matching took up to 11 minutes (on a typical desktop workstation – no NASA supercomputer here!), while clinicians took up to 130 minutes (⌛️!!) for the highest burden cases.

5/ Lesion matching is a harder task (for both 👩‍⚕️and 🤖) in subjects with higher disease burden. Here are IRV metrics as a function of lesion count – plots look similar if you compare automated to physician consensus.

4/ Automated matching met the benchmark of IRV in all study cohorts – meaning the variability between 🤖 and 👩‍⚕️ was similar to variability between 👨‍⚕️ and 👩‍⚕️

3/ In our study, two #nucmed physicians matched lesions between pairs of #PETCT images of cancer pts. Then, we used an automated 🤖approach to do the same. We assess differences between clinicians 👨‍⚕️ and differences between physician consensus 👨‍⚕️👩‍⚕️ and automated🤖

2/ Inter-reader variability (IRV) has been widely studied in image segmentation contexts, but another task clinicians 👩‍⚕️👨‍⚕️ perform is longitudinal comparison of multiple images to track lesions in time and determine change. This task is critical in determining tx. response!

1/ Excited to share our latest paper! 🧵 Performance of an automated registration-based method for longitudinal lesion matching and comparison to inter-reader variability doi.org/10.1088/1361-6560/ac… via @ioppublishing

@EmptyBowlShow @JustinMcElroy please cover these very important new cereals

I hope neither of you are adhering to an intro/methods/results/conclusion format 😂

This month has been pretty cool - I defended my PhD on imaging #immunotherapy response using #PET from ⁦@wiscmedphys⁩, and am now starting as a research scientist at ⁦@aiq_solutions⁩. Excited for new challenges! 😄

hey look its me😅

4/ Early detection+monitoring of toxicity is critical for keeping pts on #immunotherapy! Our work suggests #PETimaging can play a role here, but big limitation is when scans are acquired after tx start.

3/ Zooming out to all pts, things get messier - but general trend is elevated organ uptake at time of clinical toxicity dx (or sometimes before!). Worth noting that organ PET quantification here is automatic thanks to #AI-based organ segmentation.

2/ In an example pt, elevated bowel uptake is seen on scan2 and scan3 before clinical colitis dx 3 wks later (b,c,d). Elevated lung uptake seen on scan4 at time of clinical pneumonitis dx (e,f,g).

1/ A belated xmas present - our latest article detecting #immunotherapy toxicity on FDG PET was published in #EJNMMI. Elevated uptake in thyroid, lung, and bowel can indicate toxicity in those organs! Read the deets at: link.springer.com/article/10…

Is this how it's supposed to feel???