CENTER FOR SCALABLE DATA ANALYTICS
AND ARTIFICIAL INTELLIGENCE
NATIONAL RESEARCH DATA INFRASTRUCTURE
FOR MICROSCOPY AND BIOIMAGE ANALYSIS
How LLMs impac
BioImage Da a Science
Robe Haase
These slides can be eused unde he e ms o he CC-BY 4.0 license unless men ioned o he wise.
Funded by
h ps://doi.o g/10.5281/zenodo.17669681
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 2
The conso ium NFDI4BIOIMAGE
10/04/24
2
2
2
11 co-applican ins i u ions (6 Task A eas)
Lead ins i u ion: Hein ich Heine Uni e si y Düsseldo
Speake : P o . D . S e anie Weid kamp-Pe e s
12 pa icipa ing ins i u ions
communi y use cases
da a s ewa ds (DaS s) & esea ch so wa e enginee s
IT
IT
IT IT in as uc u e
(s o age & playg ound: @Uni Müns e & @Uni F eibu g)
Communi y-o ien ed se ices (implemen ed o planned)
•Help Desk o use communi y suppo
•T aining po olio
•Bioimage RDM playg ound in as uc u e
•Suppo ing ep oducible image analysis
Bu no : a
cen al da a a chi e o all bioimaging da a
Modi ied a e : S e anie Weid kamp-Pe e s. (2023, Juni 22). NFDI4BIOIMAGE - Na ional Resea ch Da a In as uc u e
o Mic oscopy and BioImage Analysis - Online Kick-O 2023. Zenodo. h ps://doi.o g/10.5281/zenodo.8070038
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 3
NFDI4BIOIMAGE Objec i es
310/04/24
Modi ied a e : S e anie Weid kamp-Pe e s. (2023, Juni 22). NFDI4BIOIMAGE - Na ional Resea ch Da a In as uc u e
o Mic oscopy and BioImage Analysis - Online Kick-O 2023. Zenodo. h ps://doi.o g/10.5281/zenodo.8070038
Ou o scope:
Medical, clinical, and p eclinical imaging (MRT, PET, ul asound, e c.)
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 4
Bio-image Analysis
Biologis
Domain-specialis
( ocused on
eal-wo ld p oblems)
Compu e Scien is
Me hod + in as uc u e specialis
(algo i hm-cen e ed)
Bio-image Analys
Gene alis
→
(da a-d i en,
se ice-o ien ed)
Kube ne es
clus e
Many
compu e s
wo king
oge he
T ibolium
cas aneum
Flou
bee le
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 5
Bio-image Analysis
•My job …
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 6
Bio-image Analysis
•My job is changing, since we ha e Cha GPT
P omp :
“Please segmen he
nuclei in he blue
channel o he
image”
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 7
Bio-image analysis beyond image
segmen a ion
•C ossing his bo de seems challenging
“F on ie o
pe ec ionism”
Fea u e
ex ac ion
Dimensionali y
educ ion
Classi ica ion Clus e ing
S a is ics Visualiza ion
Backg ound
emo al
Denoising
Decon olu ion
Segmen a ion
Insigh
Wide a ie y o [compu a ional] skills equi ed o achie e all his
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 8
Consul ing LLMs
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 9
Consul ing LLMs
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 26
Collabo a i e wo king wi h AI assis an s
•New AI-assis ed coding
solu ion appea
daily
•Human-human-AI
in e ac ion
•T anspa ency
•Knowledge exchange
h ps://gi hub.com/haesleinhuep /gi -bob/
h ps://www.na u e.com/a icles/s43588-025-00781-1
H
Domain
expe
H
Da a
analys
Gi hub / Gi lab CI
Hi! I ha e da a and would like o
measu e x.
I p opose o y algo i hm alg.
gi -bob commen on how o do
his wi h Py hon
AI
gi -bob
LLM se ice
p o ide 1
LLM se ice
p o ide 2
Ce ainly! T y his:
om algo i hms impo alg
x = alg(da a)
boxplo (x)
Fo esul isualiza ion le ’s be e
use a iolinplo .
gi -bob ask claude o y his!
I c ea ed a
no ebook o
using alg and
d awing a
iolinplo :
LLM se ice
p o ide 3
LLM
gp -4o
LLM
claude
LLM
gemini
Tha looks ok/co ec !
Nice! Thanks!
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 28
Summa y & Ou look
•Using AI / LLMs in [bioimage] da a analysis
seems mo e and mo e una oidable
•Main aining good scien i ic p ac ice is
challenging (esp. human-AI pee - e iew)
•Relying on ins i u ional in as uc u e
exclusi ely is impo an bu ha d
•Lea n how o use AI
esponsibly
(-> ain he aine s)
•We a e on he edge owa ds AI-agen s
analyzing image da a au onomously.
h ps://doi.o g/10.5281/zenodo.17669681
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 29
Bio-Image Analysis Communi ies
•Image/da a analysis + da a managemen suppo
•Ne wo king oppo uni ies
•T ainings
h ps://n di4bioimage.de/help-desk/ h ps://www.globias.o g/abou -
globias/globias-associa ion
h ps://image.sc/
Image Science Communi y Fo um
Na ional In as uc u e o Resea ch
Da a Managemen in BioImaging
LLMs Bio-Image Analysis, NFDI4DS Con e ence, Robe Haase, No embe 25 h 2025 30
Collabo a o s & con ibu o s
Ch is ian Tische (EMBL Heidelbe g), Jean-Ka im Hé iché
(EMBL Heidelbe g), Nico Sche (MPI CBS Leipzig), Loïc A.
Roye (CZ Biohub), Elena Nicolay (UFZ Leipzig), Ke in
Yamauchi (ETH Zu ich / Basel), Se h Hinz, Teun A.P.M.
Huijben, Jo dão B agan ini (CZ Biohub San F ancisco), llan
Theodo o (CZ Biohub and he S a e Uni e si y o Campinas,
B azil), Ian C. (Uni e si y o B i ish Columbia), P adeep
Rajasekha (WEHI Aus alia)
Acknowledgemen s
Communi ies & pla o ms Lea Kabjesz, Lea Gihlein, Ma a Lampe
RH acknowledges he inancial suppo by he Fede al Minis y o Educa ion and Resea ch o Ge many
and by Sächsische S aa sminis e ium ü Wissenscha , Kul u und Tou ismus in he p og amme Cen e o
Excellence o AI- esea ch „Cen e o Scalable Da a Analy ics and A i icial In elligence D esden/Leipzig“,
p ojec iden i ica ion numbe : ScaDS.AI
BIDS Team @ ScaDS.AI / UL
Funding
h ps://doi.o g/10.5281/zenodo.17669681
