Poster preview
ePoster
DECODING NEURON-TYPE HETEROGENEITY FROM CALCIUM ACTIVITY: A DEEP LEARNING FRAMEWORK FOR IMPROVED NEUROMODULATION ANALYSIS
Dimitris Dimitriouand 4 co-authors
Cyprus Institute of Neurology and Genetics
FENS Forum 2026 (2026)
Barcelona, Spain
Board PS05-09AM-667
Presentation
Date TBA
Board: PS05-09AM-667
Event Information
Poster Board
PS05-09AM-667
Poster
View posterAbstract
Calcium imaging is widely used to study neuronal population dynamics and neuromodulation. A critical and commonly overlooked limitation in neuromodulation research is the heterogeneity of neuron types (excitatory vs inhibitory, including different inhibitory subclasses) within recorded populations. While classification from evoked responses can yield higher accuracy, baseline activity, though subtler and more challenging to distinguish, provides greater generalizability across conditions, enabling robust neuron-type identification independent of specific stimuli. To address this, we developed a deep learning framework for automatic neuron-type classification, from baseline calcium imaging data. Using a convolutional-recurrent architecture, the model captures amplitude and temporal features of individual neuronal activity, including differences in recording duration, activation magnitude, stimulus dependence, and class imbalance.
To further enhance robustness and interpretability, we employ two complementary strategies: segmenting calcium traces into overlapping temporal windows and neuron-level aggregation across multiple activity segments. The network adopts a hierarchical two-stage classification scheme, first distinguishing excitatory from inhibitory neurons and subsequently resolving inhibitory subclasses (e.g., SST, VIP, and PV).
The model was evaluated on two independent calcium imaging datasets (visual cortex, mouse) with CRE-line GCaMP expression of Excitatory SCc17a7, VIP, SST, and PV neurons, demonstrating stable and reproducible performance (Macro-average F1 score across both datasets = 0.75-0.86). These findings highlight the importance of temporal features for distinguishing neuron types from baseline activity.
Overall, this work offers a practical, interpretable method for baseline-based neuron-type classification in neuromodulation pipelines. Prioritizing generalizability over evoked-response accuracy, it enables precise, condition-independent quantification of cell-type-specific responses, boosting study interpretability and reproducibility.
Recommended posters
HIPPIE: A MULTIMODAL DEEP LEARNING MODEL FOR ELECTROPHYSIOLOGICAL CLASSIFICATION OF NEURONS
Jesus Gonzalez Ferrer, Julian Lehrer, Jinghui Geng, Mohammed Andres Mostajo-Radji, David Haussler
A DATA-DRIVEN BIOPHYSICAL FRAMEWORK BRIDGING NEUROMODULATION TO MESOSCALE DYNAMICS
Enia Ardid, Kaleb Belay, Javier Alegre Cortes, Nikolas Karalis
COMPREHENSIVE SINGLE-CELL IDENTIFICATION OF EXCITATORY AND INHIBITORY NEURONS IN CULTURED NEURAL NETWORKS USING HIGH-DENSITY MICROELECTRODE ARRAYS
Yusaku Yamagishi, Masato Sugino, Kenta Shimba, Kiyoshi Kotani
A HIGH-THROUGHPUT SYNAPTIC CALCIUM IMAGING ANALYSIS PIPELINE FOR SCREENING NMDAR MODULATORS
John Carl Begley, Paul Turko, Camin Dean
HIGH-DENSITY SPATIOTEMPORAL SINGLE-UNIT ACTIVITY PROFILING AND CELL-TYPE IDENTIFICATION IN THE HUMAN NEOCORTEX
Orsolya Farkas, William Munoz, Richard Hardstone, Brian Coughlin, Irene Caprara, Mohsen Jamali, István Ulbert, Ziv M. Williams, Sydney S. Cash, Angelique C. Paulk, Domokos Meszéna
FUNCTIONAL AND STRUCTURAL IMAGING OF MULTIPLE INDIVIDUAL PYRAMIDAL NEURONS IN FREELY BEHAVING MICE OVER SEVERAL WEEKS
Marios Koutsoftas, Matthias Klumpp, Stefan Vintila, Andreas Draguhn, Martin Both