Item 10 — Selection of data subsets
State whether investigators selected subsets of raw extracted data during preprocessing. For example, describe whether investigators selected a subset of the images, cropped portions of images, or extracted segments of a report. If this process is automated, describe the tools and parameters used. If performed manually, describe the training of the personnel and criteria used in their instruction. Justify how this manual step would be accommodated in context of the clinical or scientific problem, describing methods of scaling processes, when applicable.
Explanation and Elaboration: Some models may require a preprocessing step that crops or excerpts the data of interest, for instance cropping around a lesion or extracting the impression section of a report. Readers need to know how this is accomplished for reproducibility and to understand the applicability of the model. This is particularly important if the process is manual, as that manual step would have to be part of any clinical deployment of the model; evaluating the potential utility of the model requires information about how laborious the process is and what skill level is required.
| Examples |
| Cortical reconstruction and volumetric segmentation were performed with the cross-sectional pipeline of FreeSurfer, version 7.1.1 …, to extract volumes of 68 cortical regions in the Desikan-Killiany atlas and 14 subcortical brain regions as well as total cerebrospinal fluid volume, total gray matter volume, and total brain volume with and without ventricles. Figure S1 lists all features derived from FreeSurfer. Image quality metrics were estimated using the MRI Quality Control (MRIQC) tool, version 0.16.1 …. Automatic quality control of the FreeSurfer segmentations was performed using the Euler number, where outliers, defined as 1.5 × IQR below the first quartile …, for each scanner were excluded from our experiments. [Materials and methods section (6), CC-BY 4.0] |
| During image preprocessing, all patient identification information and the peripheral areas in the USG images were cropped out. We identified the coordinates of fan-shaped USG region by ‘Sequence of Ultrasound Regions’ DICOM header, in order to ensure that the cropped image contained only the fan-shaped USG region where annotations and dimension measurements had been cropped out. We also removed markers which were made by sonographers in some images (S1 Appendix in S1 File). The images were then resized to 1333 pixels wide and 800 pixels tall before inputted into the CNN. [Materials and methods (27), CC-BY 4.0] |
| The tumor segmentation process on MRI is shown in Fig.2. Tumor delineation was performed on the entire three-dimensional tumor volume on T2WI images by a radiologist (the first author) using AK software (Artificial Intelligence Kit, version 3.3.0, GE Healthcare). For the segmentation of peritumoral regions, a peritumoral 3 mm region was obtained with automated dilation of the tumor boundaries by 2 mm on the outside and shrinkage of the tumor boundaries by 1 mm on the inside, resulting in a ring with a thickness of 3 mm …. The peritumoral-mesorectal fat (MRF) region was obtained by drawing along the mesorectal fascia. Thirty patients’ data were randomly selected for assessing interobserver and intraobserver agreement of feature extraction by intraclass correlation coefficient (ICC) analysis. First, a radiologist delineated the tumor volume of interest on T2WI images and repeated this process after 3 weeks to calculate the intraobserver ICC. Two radiologists independently delineated the tumor volume of interest on T2WI images, and the interobserver ICC was calculated. Only radiomics features demonstrating an ICC greater than 0.75 were retained for further analysis. [Materials and methods (28), CC-BY 4.0] |