Table 3 Summary of performance metrics on test samples, for both SEM and TEM models.

Modality	Model	Test sample(s)	Axon Dice similarity	Myelin Dice similarity	Pixel-wise accuracy	Sensitivity	Precision
SEM	Trained on rat samples	Rat 1	0.9089	0.8193	0.8510	0.9699	0.8468
		Rat 2	0.9244	0.8389	0.8822	0.9876	0.7987
		Human	0.8089	0.7629	0.8114	0.9300	0.7306
TEM	Trained on mice samples	Mice	0.9493	0.8552	0.9451	0.9597	0.9647
TEM	Trained on mice samples	Macaque	0.9069	0.7519	0.8438	0.9429	0.8129

The SEM model was trained on rat spinal cord samples, and evaluated on rat and human spinal cord samples, while the TEM model was trained on mice brain samples, and evaluated on mice and macaque brain samples. For each sample, axon Dice, myelin Dice, pixel-wise accuracy, sensitivity and precision were computed. Axon and myelin Dice measure the similarity between the axon/myelin segmentation masks and the ground truth. Pixel-wise accuracy is a measure of the ratio of correctly classified pixels. Sensitivity and precision values are an indication of the capability to detect true axonal fibers and to avoid segmentation of false axonal fibers. Note that for the mice, 24 samples of the same size were used: performance metrics shown are means between all samples.

Quick links

Search