New morphological features based on the Sholl analysis for automatic classification of traced neurons

Area of Science:

• Neuroscience
• Computational Biology
• Bioinformatics

Background:

• Automatic classification of neurons is crucial for understanding neural circuits.
• Existing methods, like L-Measure software, have limitations in capturing comprehensive morphological features.
• Accurate neuron classification aids in distinguishing cell types and their functions.

Purpose of the Study:

• To extend the capabilities of L-Measure software for automatic neuron classification.
• To develop and validate novel morphological features for improved neuron categorization.
• To enhance the accuracy and discriminatory power of computational neuroscience tools.

Main Methods:

• Development of new morphological features based on modified Sholl analysis, incorporating compartment lengths and volumes.
• Utilized feature selection methods (FSM) to identify optimal feature subsets.
• Employed supervised classification tasks, including Random Forest (RF) classifiers, to evaluate feature set performance.
• Tested on mouse pyramidal and GABAergic interneurons from cortical layers 4 and 5.

Main Results:

• The Random Forest classifier, combined with a Wrapper method, demonstrated superior performance.
• Feature subsets like U-WNAD and U-LM-WNAD significantly outperformed existing methods (WN, A, D, LM).
• The U-LM-WNAD set achieved the highest Area Under the Curve (AUC) values across all tested FSM.
• New features exhibited high discriminatory power, exceeding those from L-Measure alone, with improved AUC and accuracy (Acc).

Conclusions:

• The newly proposed morphological features possess significant discriminatory power for automatic neuron classification.
• The U-WNAD and U-LM-WNAD feature sets yielded the highest AUC and Acc values, confirming their effectiveness.
• These advancements enhance the ability to classify neurons based on morphology, extending L-Measure's utility.