Towards a Pseudo-Labeling Workflow for Celltype-Classification from Explanted Brain Slice Recordings

2026-07-09 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, medium

Summary

This paper introduces an unsupervised machine learning workflow designed to pseudo-label extracellular spikes from human brain slice Microelectrode Array (MEA) recordings, classifying them into pyramidal cells and interneurons. The proposed pipeline processes raw data through bandpass filtering, threshold-based spike detection, frame alignment, and normalization. For classification, it employs dimensionality reduction techniques like PCA, t-SNE, and UMAP, combined with clustering algorithms such as GMM and k-means. The workflow also explores template matching and OSort under different curation strictness levels to assess online system feasibility. Evaluated using within-cluster Pearson correlation, Silhouette score, and Calinski–Harabasz index, the study demonstrates that stricter data curation significantly enhances cluster separation, albeit with some reduction in data inclusivity. For instance, OSort's "strict" setting achieved a Silhouette score of 0.478 and a CH index of 5.25k, with a similarity of 0.838.

Key takeaway

For Machine Learning Engineers developing real-time cell-type classifiers from brain slice MEA recordings, you should prioritize data curation strictness to achieve robust cluster separation. Employing methods like OSort with a "strict" curation approach significantly improves inter-cluster distinction, as evidenced by a Silhouette score of 0.478. Additionally, always estimate the optimal cluster count to prevent merging sub-populations and ensure stable pseudo-labels for downstream applications.

Key insights

Unsupervised ML workflow pseudo-labels brain slice MEA spikes into cell types, balancing curation strictness with cluster quality.

Principles

• Stricter data curation improves inter-cluster separation.
• Inclusivity can boost within-cluster similarity.
• Optimal cluster count should be estimated.

Method

The workflow involves bandpass filtering, threshold-based spike detection, frame alignment, and min-max normalization. Then, dimensionality reduction (PCA, t-SNE, UMAP) and clustering (GMM, k-Means) are applied, followed by optional cleaning or OSort curation.

In practice

• Apply OSort with "strict" curation for best separation.
• Estimate optimal cluster count for better separability.
• Use min-max normalization for model generalization.

Topics

• Pseudo-labeling
• Cell-type Classification
• Microelectrode Arrays
• Unsupervised Machine Learning
• Spike Sorting
• Data Curation
• OSort

Code references

• es-ude/denspp.offline

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Review of Pseudo-Labeling for Computer Vision — Pseudo-labeling extends beyond semi-supervised learning to enhance self-supervised and unsupervised methods.
• External Clustering Validation by the Homogeneity-Parsimony Trade-off — A new framework quantifies the homogeneity-parsimony trade-off in external clustering validation using information-theoretic scores, unifying evaluation criteria.
• When Does Consensus Beat Voting? A Critical Analysis of Statistical Label Fusion in Medical Image Segmentation — STAPLE often underperforms in medical image segmentation, making robust majority voting and deep consensus models with image data more effective.
• Uncovering smooth structures in single-cell data with PCS-guided neighbor embeddings — NESS improves single-cell data visualization by using stability measures to correct distortions in neighbor embedding algorithms.
• Converge to Surprise: Evolutionary Self-supervised Image Clustering — Self-supervised image clustering can achieve state-of-the-art results by maximizing a "surprise score" without per-step loss targets, using an evolutionary strategy.
• A Data-Centric Framework for Intraoperative Fluorescence Lifetime Imaging for Glioma Surgical Guidance — A data-centric AI framework improves FLIm accuracy for glioma surgery by refining labels and enhancing model robustness.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.