Record Details

A phantom lesion dataset for precise estimation of lesion volume from ultrasound images

Harvard Dataverse (Africa Rice Center, Bioversity International, CCAFS, CIAT, IFPRI, IRRI and WorldFish)

View Archive Info
 
 
Field Value
 
Title A phantom lesion dataset for precise estimation of lesion volume from ultrasound images
 
Identifier https://doi.org/10.7910/DVN/OSYZUH
 
Creator Wijata, Agata M.
Biesok, Marta
Juszczyk, Jan
Nalepa, Jakub
 
Publisher Harvard Dataverse
 
Description This dataset is designed for the evaluation of methods for estimating the volume of lesions using in vivo approaches. Two tissue phantoms: (a) pork loin and (b) turkey meat (poultry) were exploited to capture the ultrasound images in controlled lesion-tissue settings. Inside each of the phantoms, a hydrogel ball imitating a lesion was placed in 16 positions within the tissue phantom (16 for pork and 16 for turkey, totaling to 32 independent configurations). The hydrogel absorbs water, making the structure made of it precisely visible in ultrasound imaging. Due to the high water content, the hydrogel ball (a phantom lesion) is presented in the form of a homogeneous object with low intensity value. Additionally, the spherical shape allows for performing precise reference measurements using the ex vivo method. Each phantom was imaged using a Philips iU22 US scanner and two transducers: (i) the linear probe Philips L12-5 and (ii) the convex probe Philips X6-1. The position of the linear probe was tracked using the Polaris Vicra optical system (Northern Digital Inc.), and the image was captured using an S-Video to USB converter (Imaging Source DFG/USB2pro).



Overall, we collected 32 sets of images and measurements from two different types of meat (16 independent phantom lesion placement configurations for pork and turkey, respectively). Each set consists of 2D and 3D images, and the result of a caliper measurement. Reconstruction of the 3D volume from tracked 2D images was carried out using the Public software Library for UltraSound (PLUS) toolkit, and the expert delineations were prepared by a reader (5 years of relevant experience) using a dedicated module of 3D Slicer.



The structure of this dataset is as follows (note that the same directory structure is followed for both “poultry” and “pork” phantoms):



- track

-- original – this directory contains the original ultrasound images (in the MHA format) accompanied with the corresponding spatial location captured while scanning the phantom; this directory additionally contains an XML configuration file with the translation and rotation matrices which allow for determining the position of the image relative to the tracked marker)

-- reconstruction – this directory contains the reconstructed 3D ultrasound images in the MHA format (the reconstruction was performed based on the corresponding 2D images with appropriate spatial information)

-- label – this directory contains the ground-truth delineations (MHA) of the lesion (the annotations were performed in the reconstructed 3D images)



- convex

-- original – this directory contains the original (DICOM) 3D ultrasound images captured using a convex probe

-- mha – this directory contains the original 3D ultrasound images captured using a convex probe and converted to MHA files

-- label – this directory contains the ground-truth delineations of the lesion (the annotations were performed in the MHA files)



- screen – this directory contains the png files with the annotated manual measurements of the length and width of the lesion
 
Subject Engineering
Mathematical Sciences
Medicine, Health and Life Sciences
 
Date 2023-12-29
 
Contributor Wijata, Agata M.