Deep learning-based real-time detection of neurons in brain slices for in vitro physiology

Abstract

A common electrophysiology technique used in neuroscience is patch clamp: a method in which a glass pipette electrode facilitates single cell electrical recordings from neurons. Typically, patch clamp is done manually in which an electrophysiologist views a brain slice under a microscope, visually selects a neuron to patch, and moves the pipette into close proximity to the cell to break through and seal its membrane. While recent advances in the field of patch clamping have enabled partial automation, the task of detecting a healthy neuronal soma in acute brain tissue slices is still a critical step that is commonly done manually, often presenting challenges for novices in electrophysiology. To overcome this obstacle and progress towards full automation of patch clamp, we combined the differential interference microscopy optical technique with an object detection-based convolutional neural network (CNN) to detect healthy neurons in acute slice. Utilizing the YOLOv3 convolutional neural network architecture, we achieved a 98% reduction in training times to 18 min, compared to previously published attempts. We also compared networks trained on unaltered and enhanced images, achieving up to 77% and 72% mean average precision, respectively. This novel, deep learning-based method accomplishes automated neuronal detection in brain slice at 18 frames per second with a small data set of 1138 annotated neurons, rapid training time, and high precision. Lastly, we verified the health of the identified neurons with a patch clamp experiment where the average access resistance was 29.25 M[Formula: see text] (n = 9). The addition of this technology during live-cell imaging for patch clamp experiments can not only improve manual patch clamping by reducing the neuroscience expertise required to select healthy cells, but also help achieve full automation of patch clamping by nominating cells without human assistance.

Figure 1

After initial (a) training and validation using annotated input images, testing (b) shows a successful detection of neurons in unannotated, unaltered images.

Figure 2

(a) Representative example of unaltered (top) and enhanced (bottom) images of acute slice under DIC. (b) Representative plot of F1 score vs confidence threshold, demonstrating peak in F1 score at a confidence threshold of 0.3. (c) left Relationship between precision and recall for the enhanced network tested on enhanced and unaltered data set test images. right Relationship between precision and recall for the unaltered network tested on enhanced and unaltered data set test images. (d) Summary of mean average precision of both networks for both enhanced and unaltered inputs. (e) Summary of F1 score of unaltered and enhanced networks for both enhanced and unaltered inputs. $\text{Scale bar}=10\mathrm{\mu }$m.

Figure 3

(a) Convergence on training and validation loss with respect to number of epochs. Black lines represent the unaltered trained model losses, and gray represents the enhanced trained model losses. Solid lines represent training loss, and dashed lines represent validation loss. (b) The bar chart shows mean ± SD comparison of the average accuracy between the unaltered net and enhanced net on the unaltered data set test images. A student’s t-test ($\alpha$ = 0.05) acknowledges that the difference between the means is statistically significant; t(36)=5.12, p$<0.001$. (c) Box plot comparison of the confidence scores distribution for unaltered and enhanced networks tested on the unaltered data set test images. The notches represent the confidence interval around the median using a Gaussian-based asymptotic approximation. The ends of the boxes are at the first and third quartiles while the whiskers represent the minimum and maximum confidence scores. (d,e) Example of both networks identifying neurons in a test image. left initial prediction (red) of neurons. right bounding boxes for annotation (blue), correct prediction (true positive—green), incorrect prediction (false positive—red), and undetected neurons (false negative—pink). Scale bar: $10\mathrm{\mu }$m.

Figure 4

(a) Image of a network-identified neuron in patch clamp whole-cell configuration. The blue bounding boxes indicate identified neurons. The numbers ranging from 0 to 1 indicate the network’s confidence that the box contains a neuron. The pipette recording electrode is visible on the lower left quadrant resting on the leftmost of the three identified neurons. (b) Distribution of access resistance indicate that 8 out of 9 cells (89%) yielded high quality whole cell recordings. The white line indicates the median (21.7), the box width indicates the interquartile range (9.6), and the whiskers indicate the range of the data, excluding outliers. (c) Representative current clamp trace and (d) voltage clamp trace from a neural network-identified neuron in whole-cell configuration. Scale bar: $10\mathrm{\mu }$m.