Researchers have developed an automated preprocessing workflow designed to standardise computed tomography imaging data for machine learning and deep learning applications within forensic anthropology. The investigation addresses critical operational hurdles in forensic science, where the adoption of medical computed tomography scans has grown to help overcome ethical, logistical, and physical preservation difficulties associated with traditional cadaveric collections. However, deploying artificial intelligence on large collections of medical imaging data continues to present substantial challenges, specifically regarding data privacy safeguards and technical standardisation across imaging repositories.
To establish a consistent preparation method for forensic anthropology research and education, the study team created an end-to-end pipeline programmed in Python. The resulting framework organizes image preparation into three distinct technical stages: the de-identification and sorting of patient data, the computational removal of the examination table, and specialized bone windowing. The workflow was specifically constructed for and applied to computed tomography scans of the thorax, using ground-truth reference models alongside standard performance metrics to evaluate system outcomes.
The initial stage of the automated pipeline focuses on privacy protection by sanitising medical records through a dedicated software script. During preprocessing validation, the de-identification procedure addressed metadata across two hundred examined Digital Imaging and Communications in Medicine files. The script successfully processed all predefined identifying attributes contained within the datasets, establishing verified privacy handling before subsequent computational training and analytical procedures were conducted.
For the physical segmentation tasks, the researchers implemented a custom U-Net architecture tailored to handle both the extraction of the patient examination table and the subsequent bone windowing. The custom network architecture was applied across the thoracic imaging series to standardize the inputs.
The patient table removal network successfully eliminated the examination table in 96 percent of the evaluated datasets, representing 192 out of the 200 scan series. Furthermore, the model attained an accuracy rating of 99.94 percent when tested on original, non-augmented test images, demonstrating consistent structural recognition across the imaging sample.
The bone segmentation model achieved a Dice score of 93.12 percent alongside a precision of approximately 90 percent. When the investigators generated segmented three-dimensional reconstructions from these processed scans, the resulting models exhibited high agreement with the designated ground-truth reference. Technical comparisons confirmed that the segmented outputs showed minimal geometric deviation from the corresponding reference standard.
The proposed integrated workflow provides an ethical, standardized, and efficient solution for de-identifying and preprocessing computed tomography imaging data. By enabling the preparation of a scalable dataset for artificial intelligence applications, the methodology supports analytical development for modern populations. The authors noted that future research will pursue expansion of the workflow to full-body computed tomography scans and additional medical imaging modalities.
Subject of Research: Automated CT image preprocessing workflow for forensic anthropology AI
Article Title: Development and technical validation of a standardised CT image preprocessing workflow for reproducible AI for forensic anthropological applications
Article References: Jani, G., Patel, S., Roy, M., Sehrawat, A., & Patel, B. M. (2026). Development and technical validation of a standardised CT image preprocessing workflow for reproducible AI for forensic anthropological applications. International Journal of Legal Medicine. https://doi.org/10.1007/s00414-026-04076-6
Image Credits: AI Generated
DOI: 10.1007/s00414-026-04076-6
Keywords: computed tomography, forensic anthropology, image preprocessing, DICOM de-identification, patient table removal, bone windowing, U-Net architecture, deep learning, machine learning, thorax CT, Dice score, bone segmentation, 3D model reconstruction, metadata anonymisation, cadaveric collections, reproducible artificial intelligence
News Source: Blake Davidson. (November 17, 2026). Automated Preprocessing Pipeline Standardises CT Imaging Data for Forensic Artificial Intelligence. Scienmag.



