Improving a cortical pyramidal neuron model's classification performance on a real-world ecg dataset by extending...

Area of Science:

• Computational neuroscience
• Biophysics

Background:

• Pyramidal neurons possess complex structures and active conductances enabling nonlinear dendritic computation.
• Understanding neuronal computation for real-world data classification is a growing area of research.

Purpose of the Study:

• To investigate the classification capabilities of a detailed pyramidal neuron model using electrocardiogram (ECG) data.
• To evaluate the impact of input mirroring on the neuron's classification performance.

Main Methods:

• Applied a detailed pyramidal neuron model and the perceptron learning algorithm to classify ECG signals.
• Utilized Gray coding to convert ECG signals into spike patterns.
• Assessed classification performance across different subcellular regions of the pyramidal neuron.

Main Results:

• The pyramidal neuron model initially performed poorly compared to a single-layer perceptron due to weight constraints.
• A proposed input mirroring technique substantially enhanced the neuron's classification accuracy.
• The mirroring approach demonstrated performance comparable to unconstrained learning algorithms.

Conclusions:

• Pyramidal neurons are capable of classifying complex, real-world data like ECG signals.
• Input mirroring is an effective strategy to improve the computational performance of pyramidal neuron models.
• The study highlights the potential of biophysically realistic neuron models for data classification tasks.