Pipeline outputs for the thesis "Application of radiomic feature maps for the analysis of adrenal lesions in computed tomography scans"

This record accompanies a master's thesis project on exploratory and semi-quantitative analysis of adrenal glands in chest CT. It contains the manual candidate ROIs required to reproduce the experiments (except the original CT volumes) and a full copy of the generated outputs. Original CT volumes are not included. They belong to the CT-RATE dataset (Hamamci et al.) and cannot be redistributed under that dataset's licence. Obtain the listed cases from Hugging Face: https://huggingface.co/datasets/ibrahimhamamci/CT-RATE Place the downloaded volumes in final/input/ as <case_id>.nii.gz. The case list is defined in pipeline.ps1 in the companion repository: https://github.com/ochnios/msc-proj. A helper script, utils/download_ct_rate.py, is provided there, but it may stop working if the dataset layout or API changes. What is included - final/input/manual/: manually drawn candidate ROIs for suspicious cases. These are required for contrast evaluation and suspicious-case visualisations. - final/output/: TotalSegmentator adrenal masks, voxel-wise radiomic feature maps, visualisation boards, and contrast_eval_results.csv. To rerun the experiments, provide the original CT-RATE volumes and final/input/manual/. Everything in final/output/ can be regenerated with pipeline.ps1; the included copy is provided for convenience. Case selection Cases were preselected from CT-RATE radiology reports for exploratory and semi-quantitative analysis. They do not form a formal clinical cohort or a representative research sample; results should be interpreted as case studies for preliminary assessment of the approach. Reports were searched for terms referring to adrenal lesions, including adrenal adenoma/myelolipoma and pheochromocytoma. Matching studies were reviewed, preferring reports with information on laterality, location, size, and lesion type. Studies were excluded when the report did not allow unambiguous localisation of the described region, when lesions were outside the scope of this work, or when automatic adrenal segmentation did not cover the described region sufficiently for further analysis. Ten CT volumes were retained as suspicious cases. For each suspicious case, an approximate region corresponding to the reported lesion location was drawn manually in 3D Slicer on the original CT, with the automatically segmented adrenal mask overlaid. Radiomic feature maps were not displayed during this step. These regions are not expert lesion segmentations. They are used only to compare local HU and selected radiomic features with the remaining automatically segmented adrenal. They were not independently verified by an expert, so results should be interpreted with caution. Control cases were selected from studies whose reports stated that the adrenal glands were normal. This is a weak label derived from the report, not independently verified confirmation of no lesions. Controls were used only for qualitative assessment; no lesion regions were drawn for semi-quantitative analysis. Two CT volumes were retained as external controls: train_2240_a_2 and train_8275_a_1. The contralateral adrenal of each suspicious case was also used as a within-subject control. Suspicious cases (n = 10) - train_1345_a_2 — left - train_2344_a_2 — left - train_3798_a_2 — right - train_4783_a_2 — right - train_5260_a_2 — left - train_7928_a_2 — left - train_8173_a_1 — right - train_9278_c_2 — left - train_10112_a_2 — left - train_19401_a_1 — right TotalSegmentator labels: right adrenal = 8; left adrenal = 9. Users must comply with the CT-RATE terms when obtaining the original volumes and cite CT-RATE (Hamamci et al.) in any work that uses these cases.