The DL_Track package provides an easy to use graphical user interface (GUI) for deep learning based analysis of muscle architectural parameters from longitudinal ultrasonography images of human lower limb muscles. Please take a look at our documentation for more information. This code is based on a previously published algorithm and replaces it. We have extended the functionalities of the previously proposed code. The previous code will not be updated and future updates will be included in this repository.
For detailled information about installaion of the DL_Track python package we refer you to our documentation. There you will finde guidelines not only for the installation procedure of DL_Track, but also concerding conda and GPU setup.
Once installed, DL_Track can be started from the command prompt with the respective environment activated:
(DL_Track) C:/User/Desktop/ python -m DL_Track
In case you have downloaded the executable, simply double-click the DL_Track icon.
Regardless of the used method, the GUI should open. For detailed the desciption of our GUI as well as usage examples, please take a look at the user instruction.
We have not yet integrated unit testing for DL_Track. Nonetheless, we have provided instructions to objectively test whether DL_Track, once installed, is functionable. Do perform the testing procedures yourself, check out the test instructions.
In order to see the detailled scope and description of the modules and functions included in the DL_Track package, you can do so either directly in the code, or in the Documentation section of our online documentation.
The previously published algorithm was developed with the aim to compare the performance of the trained deep learning models with manual analysis of muscle fascicle length, muscle fascicle pennation angle and muscle thickness. The results were presented in a published preprint. The results demonstrated in the article described the DL_Track algorithm to be comparable with manual analysis of muscle fascicle length, muscle fascicle pennation angle and muscle thickness in ultrasonography images as well as videos. The results are briefly illustrated in the figures below.
Bland-Altman plots of the results obtained with our approach versus the results of manual analyses by the authors (mean of all 3). Results are shown for muscle fascicle length (A), pennation angle (B), and muscle thickness (C). For these plots, only the median fascicle values from the deep learning approach were used, and thickness was computed from the centre of the image. Solid and dotted lines depict bias and 95% limits of agreement, respectively.
A comparison of fascicle lengths computed using DL_Track with those from UltraTrack(Farris & Lichtwark, 2016, DOI:10.1016/j.cmpb.2016.02.016), a semi-automated method of identifying muscle fascicles. Each row shows trials from a particular task (3 examples per task from different individuals, shown in separate columns). For DL_Track, the length of each individual fascicle detected in every frame is denoted by a gray dot. Solid black lines denote the mean length of all detected fascicles by DL_Track. Red dashed lines show the results of tracking a single fascicle with Ultratrack.
Wheter you want to contribute, report a bug or have troubles with the DL_Track package, take a look at the provided instructions how to best do so. You can also contact us via email at [email protected], but we would prefer you to open a discussion as described in the instructions.