R: Loads npy files to RIA image format

load_npy {RIA}

R Documentation

Loads npy files to RIA image format

Description

Loads numpy arrays from python to a RIA_image object using the reticulate package. Requires python and numpy to be installed! RIA_image is a list with three mandatory attributes.

RIA_data is a RIA_data object, which has two potential slots. $orig contains the original image after loading $modif contains the image that has been modified using functions.
RIA_header is a RIA_header object, which is list of header information.
RIA_log is a RIA_log object, which is a list updated by RIA functions and acts as a log and possible input for some functions.

Further attributes may also be added by RIA functions.

Usage

load_npy(
  filename,
  mask_filename = NULL,
  keep_mask_values = 1,
  switch_z = FALSE,
  crop_in = TRUE,
  replace_in = TRUE,
  center_in = FALSE,
  zero_value = NULL,
  min_to = -1024,
  PixelSpacing = 1,
  SpacingBetweenSlices = 1,
  verbose_in = TRUE,
  ...
)

Arguments

`filename`	string, file path to npy file.
`mask_filename`	string vector, file path to npy file of mask image. If multiple are supplied, then those voxels are kept which have one of the values of keep_mask_values in any of the supplied masks.
`keep_mask_values`	integer vector or string, indicates which value or values of the mask image to use as indicator to identify voxels wished to be processed. Usually 1-s indicate voxels wished to be processed. However, one mask image might contain several segmentations, in which case supplying several integers is allowed. Furthermore, if the same string is supplied to filename and mask_filename, then the integers in keep_mask_values are used to specify which voxel values to analyze. This way the provided image can be segmented to specific components. For example, if you wish to analyze only the low-density non-calcified component of coronary plaques, then keep_mask_values can specify this by setting it to: -100:30. If a single string is provided, then each element of the mask will be examined against the statement in the string. For example, if '>0.5' is provided i.e. the mask is probabilities after a DL algorithm, then all voxels with values >0.5 in the mask image will be kept. This can be a complex logical expression. The data on which the expression is executed is called data or data_mask, depending on whether you wish to filter the original image, that is the original image is supplied as a mask, or if you have unique mask files respectively. Therefore for complex logical expressions you can define for example: '>-100 & data<30' to consider data values between -100 and 30, or '>0.5 & data_mask<0.75' to select voxels based-on mask values between 0.5 and 0.75 for example if they represent a probability mask.
`switch_z`	logical, indicating whether to change the orientation of the images in the Z axis. Some software reverse the order of the manipulated image in the Z axis, and therefore the images of the mask image need to be reversed.
`crop_in`	logical, indicating whether to crop RIA_image to smallest bounding box.
`replace_in`	logical, whether to replace smallest values indicated by zero_value, which are considered to indicate no signal, to NA.
`center_in`	logical, whether to shift data so smallest value is equal to min_to input parameter.
`zero_value`	integer, indicating voxels values which are considered not to have any information. If left empty, then the smallest HU value in the image will be used, if replace_in is TRUE.
`min_to`	integer, value to which data is shifted to if center_in is TRUE.
`PixelSpacing`	numerical, Pixel spacing value of image.
`SpacingBetweenSlices`	numerical, Spacing between the slices value of the image.
`verbose_in`	logical, indicating whether to print detailed information. Most prints can also be suppressed using the `suppressMessages` function.
`...`	additional arguments to numpy.load.

Details

load_npy is used to transform numpy array datasets into the RIA environment. RIA_image object was developed to facilitate and simplify radiomics calculations by keeping all necessary information in one place.

RIA_data stores the numpy image that is converted to numerical 3D arrays using the reticulate package. The function stores the original loaded image in RIA_data$orig, while all modified images are stored in RIA_data$modif. By default, the original image RIA_data$orig is untouched by functions other than those operating in load_npy. While other functions operate on the RIA_data$modif image by default.
Due to memory concerns, there can only be one RIA_data$orig and RIA_data$modif image present at one time in a RIA_image. Therefore, if image manipulations are performed, then the RIA_data$modif will be overwritten. However, functions can save images into new slots of RIA_image, for example discretized images can be saved to the discretized slot of RIA_image.
load_npy not only loads the image, but also can perform minimal manipulations on the image itself.
crop_in logical variable is used to indicate, whether to crop the image to the smallest bounding box still containing all the information. If TRUE, then all X, Y and potentially Z slices containing no information will be removed. This allows significant reduction of necessary memory to store image data.
zero_value parameter is used to indicate HU values which contain no information. If left empty, then the smallest value will be considered as indicating voxels without a signal.
replace_in logical can be used to change values that are considered to have no signal to NA. This is necessary to receive proper statistical values later on.
center_in logical is used to indicate whether the values should be shifted. Some vendors save HU values as positive integers to spare memory and minimalize file sizes. Therefore, in some instances shift of the scale is needed. By default, the values are shifted by -1024, but in other cases a different constant might be required, which can be set using the min_to input.

RIA_header is a list containing the most basic patient and examination information present in the npy file. Data is limited to the pixel spacing and spacing between the slices information.

RIA_log is a list of variables, which give an overview of what has been done with the image. If the whole RIA_image is supplied to a function, the information regarding the manipulations are written into the $events array in chronological order. Furthermore, some additional information is also saved in the log, which might be needed for further analysis.

Value

Returns a RIA_image object. RIA_image is a list with three mandatory attributes.

RIA_data is a RIA_data object containing the image in $orig slot.
RIA_header is a RIA_header object, which is s list of header information.
RIA_log is a RIA_log object, which is a list updated by RIA functions and acts as a log and possible input for some functions.

References

Márton KOLOSSVÁRY et al. Radiomic Features Are Superior to Conventional Quantitative Computed Tomographic Metrics to Identify Coronary Plaques With Napkin-Ring Sign Circulation: Cardiovascular Imaging (2017). DOI: 10.1161/circimaging.117.006843 https://pubmed.ncbi.nlm.nih.gov/29233836/

Márton KOLOSSVÁRY et al. Cardiac Computed Tomography Radiomics: A Comprehensive Review on Radiomic Techniques. Journal of Thoracic Imaging (2018). DOI: 10.1097/RTI.0000000000000268 https://pubmed.ncbi.nlm.nih.gov/28346329/

Examples

## Not run: 
 #Image will be croped to smallest bounding box, and smallest values will be changed to NA
 RIA_image <- load_npy("/Users/Test/Documents/Radiomics/John_Smith/npy_folder/sample.npy")
 
## End(Not run)

[Package RIA version 1.7.2 Index]