Understanding visual processing at the retinal level is crucial for identifying the foundational elements of the biological visual pathway. Characterizing the diversity of Retinal ganglion cells (RGCs) provides significant insight into this process, as they integrate primary light encoding and transmit it to the brain as feature extractor channels, with each channel corresponding to a specific type of RGC. Despite numerous classification methods proposed over the years, characterizing the diversity of RGCs remains a challenge. Existing methods either fall short in capturing the entire spectrum of cell types or demand serial and resource-intensive protocols, hindering their utilization in large-scale recordings. To address these challenges, our study proposes an end-to-end computational pipeline that reconciles single-cell analysis with large-scale characterization via simulation. By employing models of cell functional behavior, we avoid serial stimulation, and streamline data generation for classification through prediction. With this predictive capability, we employ simulation to apply a comprehensive, well-established classification framework based on sequential analysis to multi-electrode array recordings. To train the models with time-optimized recordings, we employ data from a novel stimulus, the Multiple Spatial Frequency (MSF) checkerboard. This stimulus retains the desirable features of the standard noisy approaches while widening the range of spatial frequencies used for retinal stimulation. Our approach allows for simultaneous observation of spatial integration dynamics across all cells observed in high-throughput techniques. The results of this study demonstrate the MSF stimulus's effectiveness in capturing spatial integration dynamics, highlighting its importance in characterizing retinal output channel features in large-scale recordings, including surround suppression, alongside established properties. Moreover, our findings validate the effectiveness of the proposed computational framework in generating reliable predictions for task-transfer scenarios. In summary, our study presents a concrete method for RGC classification that integrates various cell properties, paving the way for a more thorough retinal analysis and contributing to advancements in peripheral visual encoding.
La caratterizzazione della diversità delle cellule gangliari retiniche (RGC) è fondamentale per la comprensione dell'elaborazione degli input luminosi implementata dal sistema visivo biologico. Nonostante numerosi metodi di classificazione proposti nel corso degli anni, caratterizzare la diversità delle RGC rimane una sfida. I metodi esistenti spesso non riescono a catturare l'intero spettro di tipi cellulari o richiedono protocolli seriali e dispendiosi, che risultano difficilmente applicabili su larga scala. Questo studio si propone di colmare tale lacuna mediante l'implementazione di una pipeline computazionale end-to-end che permette una classificazione completa e parallelizzata tramite simulazione. Utilizzando modelli dei pattern di attivazione delle cellule, evitiamo la stimolazione seriale e semplifichiamo la generazione di dati per la classificazione mediante previsione. Con questa capacità predittiva, impieghiamo la simulazione per applicare un framework di classificazione completo e ben consolidato basato su analisi sequenziale a registrazioni di array multi-elettrodo. Per allenare i modelli in modo ottimale, utilizziamo dati ottenuti con un nuovo approccio di stimolazione, la Multiple Spatial Frequency (MSF) checkerboard. Questo stimolo conserva le caratteristiche desiderabili delle checkerboard standard, ma al contempo estende la gamma di frequenze spaziali utilizzate per stimolare la retina. Ciò consente di osservare simultaneamente le dinamiche di integrazione spaziale su tutte le cellule osservate contemporaneamente. Con questo studio dimostriamo l'efficacia dello stimolo MSF nell'evocare dinamiche di integrazione spaziale, inclusi fenomeni come la soppressione del contorno, oltre alle proprietà consolidate. Inoltre, i risultati confermano l'efficacia della pipeline nel fornire previsioni accurate, garantendo un alto tasso di corretta classificazione. Questi due fattori insieme convalidano l'efficacia del framework proposto. In sintesi, il nostro studio presenta un metodo concreto per la classificazione delle RGC che integra diverse proprietà cellulari, aprendo la strada a un'analisi retinica più approfondita e contribuendo ai progressi nell'encoding visivo periferico.
Towards a comprehensive characterization of Retinal Ganglion Cell diversity: bridging spatial integration dynamics in large-scale recordings
Boscarino, Chiara
2022/2023
Abstract
Understanding visual processing at the retinal level is crucial for identifying the foundational elements of the biological visual pathway. Characterizing the diversity of Retinal ganglion cells (RGCs) provides significant insight into this process, as they integrate primary light encoding and transmit it to the brain as feature extractor channels, with each channel corresponding to a specific type of RGC. Despite numerous classification methods proposed over the years, characterizing the diversity of RGCs remains a challenge. Existing methods either fall short in capturing the entire spectrum of cell types or demand serial and resource-intensive protocols, hindering their utilization in large-scale recordings. To address these challenges, our study proposes an end-to-end computational pipeline that reconciles single-cell analysis with large-scale characterization via simulation. By employing models of cell functional behavior, we avoid serial stimulation, and streamline data generation for classification through prediction. With this predictive capability, we employ simulation to apply a comprehensive, well-established classification framework based on sequential analysis to multi-electrode array recordings. To train the models with time-optimized recordings, we employ data from a novel stimulus, the Multiple Spatial Frequency (MSF) checkerboard. This stimulus retains the desirable features of the standard noisy approaches while widening the range of spatial frequencies used for retinal stimulation. Our approach allows for simultaneous observation of spatial integration dynamics across all cells observed in high-throughput techniques. The results of this study demonstrate the MSF stimulus's effectiveness in capturing spatial integration dynamics, highlighting its importance in characterizing retinal output channel features in large-scale recordings, including surround suppression, alongside established properties. Moreover, our findings validate the effectiveness of the proposed computational framework in generating reliable predictions for task-transfer scenarios. In summary, our study presents a concrete method for RGC classification that integrates various cell properties, paving the way for a more thorough retinal analysis and contributing to advancements in peripheral visual encoding.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Boscarino_ExSum.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
4.59 MB
Formato
Adobe PDF
|
4.59 MB | Adobe PDF | Visualizza/Apri |
2024_04_Boscarino_ThesisReport.pdf
accessibile in internet per tutti
Descrizione: Thesis Report
Dimensione
26.04 MB
Formato
Adobe PDF
|
26.04 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/218362