PyUFF (USTB)

An implementation of USTB's ultrasound file format (UFF) in Python.

Note that this project only implements USTB's version of UFF, which is considered version 0.0.1. Multiple versions of UFF exists, for example check out uff-reader. However, if you only planning to use USTB's version of UFF then you have come to the right place :)

Installing

Run the following command in your Python virtual environment:

python -m pip install pyuff-ustb

To verify that the installation was successful, run the following command:

python -c "import pyuff_ustb; print(pyuff_ustb.__version__)"

Reading UFF files

import pyuff_ustb as pyuff

uff = pyuff.Uff(filepath)
print(uff)  # <- print the keys of the UFF file
channel_data = uff.read("channel_data")
scan = uff.read("scan")

Writing UFF files

import pyuff_ustb as pyuff
import numpy as np

scan = pyuff.LinearScan(
    x_axis=np.linspace(-10e3, 10e3, 128),
    z_axis=np.linspace(0, 80e3, 128),
)
scan.write("my_scan.uff", "scan")
# To overwrite an existing field in the file, pass overwrite=True like so:
scan.write("my_scan.uff", "scan", overwrite=True)

UFF object structure

See the modules under pyuff_ustb/objects for all implemented UFF objects. The most important ones are ChannelData and Scan.

• ChannelData contains the raw ultrasound data under the data (channel_data.data) property and other important beamforming properties such as sampling_frequency and probe setup, etc.
• Scan is primarily a container of the points that are to be beamformed.

Check out the source code under pyuff_ustb/objects/channel_data.py in your favorite code editor to get a better understanding of the UFF object structure.

Lazy loading

PyUFF strives to only load what you need in order to speed up the reading process. This is done by using lazy loaded properties. Lazy loaded values are only actually read from the file when they are used. This is contrary to eager loading where all values are automatically read from the file when the object is created. Arrays are always eagerly loaded so that you can be sure that they are NumPy arrays (in previous versions we had a LazyArray abstraction but the benefits of this were small and the complexity cost was big).

In general, all UFF object fields are of the type cached_property. When a cached_property is accessed for the first time, its code will run, and the returned value will be cached, meaning that for most PyUFF fields, values are only read from a file once. The cached_property is further split into two types:

• compulsory_property: fields that must be set in an object in order to write to a file.
• optional_property: optional fields that may be None when writing to a file.

Additionally, there is dependent_property, which is not cached nor read from a file — instead, it is calculated from other compulsory and/or optional properties in the object. All properties of the PyUFF objects are decorated with either @compulsory_property, @optional_property or @dependent_property.

To invalidate a property from the cache (in order to re-read it from the file), simply delete the property from the object. Example:

obj = uff.read("channel_data")
data = obj.data[...]  # <- This will read the data from the file
data = obj.data[...]  # <- Subsequent calls do not read from the file
del obj.data  # <- This deletes the data from the cache
data = obj.data[...]  # <- The data is read again

You can still eagerly load all values from the file by calling pyuff_ustb.eager_load on the object. Note that eager_load does not update the object (but may read and cache properties) but returns a new copy. Example:

obj = uff.read("channel_data")
# Eagerly load all values (this usually takes a few seconds, depending on the file size)
# pyuff_ustb.eager_load returns a copy of the object, leaving the original unchanged (though perhaps with cached properties).
obj = pyuff_ustb.eager_load(obj)