Troubleshooting

Improving algorithm performance

brainmapper detects cells in a two-stage process, firstly cell candidates are detected. These are cell-like objects of approximately the correct intensity and size. These cell candidates are then classified into cells and artefacts by a deep learning step.

Cell candidate detection

If the initial cell candidate detection is not performing well, then we suggest adjusting the cell detection parameters. For a better understanding of what these parameters do, it may be useful to consult the original PLOS Computational Biology paper.

Hint

In general, false positives (non-cells being detected) is generally ok, as these will be refined in the classification step.

Cell candidate classification

The classification will use a pre-trained network by default that is included with the software. This network will usually need to be retrained. For more details, please see the guide to retraining the pre-trained network.

Fixing technical problems

As brainmapper relies on a number of third party libraries, notably

• PyTorch,

• CUDA,

• cuDNN,

there may be issues while running the software. If you are having any issues, please see the following sections:

• Speeding up cellfinder
• Debugging common error messages

Anything else

If you are still having trouble, please get in touch.