Comparison of feature-based and Uniform Manifold Approximation and Projection (UMAP)-based spike sorting for neuron classification.

<p><b>(A)</b> High-pass filtering (500 Hz–2 kHz) removes low-frequency noise from extracellular signals. <b>(B)</b> Thresholding the pronounced deflections in the filtered data identifies spikes. <b>(C)</b> Windows around each spike, centered on the trough,...

Повний опис

Збережено в:
Бібліографічні деталі
Автор: Daniel Suárez-Barrera (22676654) (author)
Інші автори: Lucas Bayones (22676657) (author), Norberto Encinas-Rodríguez (22676660) (author), Sergio Parra (22676663) (author), Viktor Monroy (22676666) (author), Sebastián Pujalte (22676669) (author), Bernardo Andrade-Ortega (22676672) (author), Héctor Díaz (22676675) (author), Manuel Alvarez (3468647) (author), Antonio Zainos (22676678) (author), Alessio Franci (143351) (author), Román Rossi-Pool (22676681) (author)
Опубліковано: 2025
Предмети:
Теги: Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
Опис
Резюме:<p><b>(A)</b> High-pass filtering (500 Hz–2 kHz) removes low-frequency noise from extracellular signals. <b>(B)</b> Thresholding the pronounced deflections in the filtered data identifies spikes. <b>(C)</b> Windows around each spike, centered on the trough, are isolated to standardize waveform comparisons. <b>(D)</b> Each spike waveform becomes a point in a high-dimensional space (time vs. voltage). <b>(E)</b> Feature-based sorting relies on linear dimensionality reduction methods (e.g., principal component analysis, wavelets), yet can be thrown off by spikes from different neurons. <b>(F)</b> Clustering in this simplified feature space may misclassify neurons due to these linear constraints. <b>(G)</b> UMAP-based sorting, however, employs a nonlinear approach that preserves both local and global structure. <b>(H)</b> This UMAP projection keeps distinct clusters and neuron-specific features more intact. <b>(I)</b> Clustering in the UMAP-projected space better distinguishes individual neurons (shown as distinct colored clusters), reducing merging errors seen with feature-based methods.</p>