Artificial Intelligence-based Software for the Smart Molecular Imaging of Fingermarks

Summary

Fingermarks are a still primary form of evidence in any crime (and are particularly important in major crimes). This collection refers to a project funded by DASA UK regarding the engineering of a TRL 6 software the specific processing of crime scenes fingermark through of Matrix Assisted Laser Desorption Ionisation Mass Spectrometry Imaging (MALDI MSI). The software integrates artificial intelligence for the smart and fast provision of the highest quality fingermark images for National Database comparison and match, offering an increased likelihood of suspect identification. The Collection specifically includes: a) project background, aims, objectives and hypotheses; b) the machine learning algorithm (code) developed to assign a Scotland Yeard grade to each MALDI MS image of fingermark recovered; c) detailed information on the scripts, files, and folders used, along with the file structure and naming conventions (data level documentation (pdf). The metadata file (.xls) contains two sheets: one for the code scripts and another for the images.

Alternative title:	SMIF
Keywords:	Fingerprints, MALDI, Imaging, AI, software, biometrics
Creators:	Francese, Simona http://orcid.org/0000-0002-1381-1262 Jensen, Elias https://orcid.org/0009-0008-3765-6798 Ugail, Hassan https://orcid.org/0000-0002-3084-1797
Contributors:	Project Leader Francese, Simona http://orcid.org/0000-0002-1381-1262 Co-Investigator Hugail, Hassan https://orcid.org/0000-0002-3084-1797 Researcher Jensen, Elias https://orcid.org/0009-0008-3765-6798 Researcher Spencer, Chloe https://orcid.org/0009-0007-5934-507X Co-Investigator Broadbent, Gemma Supervisor Chew, Simona
Academic units:	Faculty of Health and Wellbeing (HWB) > Research Centres > Biomolecular Sciences Research Centre (BSRC)
Funders:	Funder Name Grant Number Funder ID Defence and Security Accelerator ACC2036403 https://doi.org/10.13039/100012339
Copyright Holders:	Sheffield Hallam University, University of Bradford
Publisher of the data:	SHU Research Data Archive (SHURDA)
Publication date:	31 March 2025
Data last accessed:	No data downloaded yet
Embargo expiry date:	1 April 2026
DOI:	http://doi.org/10.17032/shu-0000000238
SHURDA URI:	https://shurda.shu.ac.uk/id/eprint/238
Additional details Additional details Types of data: Model Collection period: From To 3 July 2023 12 April 2025 Geographic coverage: UK, South Yorkshire (Sheffield Hallam University), West Yorkshire (University of Bradford) Data collection method: Fingermarks were collected from 13 donors with the Biomolecular Sciences Research Centre at Sheffield Hallam University without any exclusion criteria. Three fingermarks were collected per donor for a total of 39 samples under full SHU ethical approval ER52762288. Fingermarks were sprayed-coated with a solution of alpha cyano 4 hydroxycinnamic acid using and HTX Technologies M3+ Sprayer (North Carolina, US)and then subjected to MALDI MS Imaging on the SELECT SERIES MALDI MRT (Waters Corporation, Wilmslow, UK). A total of 953 images have been used to train the model. Further details about the collection of the marks, the preparation and the analysis have been reported in the Study level documentation deposited within this collection. Detailed information on the model training process, including the scripts, algorithms, and training procedures, can be found in the Model Training & Inference section of the data-level documentation. Data processing and preparation activities: MALDI MS images were normalised against the total ion current (TIC) during the initial processing of the acquired data, if this pre-processing step helped to improve clarity of the ridges. The fingerprint mass images (TIC or non-TIC normalised were then shared using ZendTo with a fingerprint expert at West Yorkshire Police who assigned a Scotland Yard grade to each one of them. For training purposes a subset was used by the model, whereas for testing purposes, these scores remained blind to the AI team. The fingerprint mass images were organised in folders according to the grade a assigned and a selection was shared blind with the AI team for testing purposes. The fingermark images underwent further pre-processing and enhancement prior to using them with the model. This included techniques such as image filtering, wavelet analysis, and minutiae extraction, as detailed in the Data Pre-processing & Enhancement section of the project documentation. These pre-processing steps ensured that the images were consistently prepared for machine learning training. The images were then organized into categorized datasets (0-5) based on grading provided by West Yorkshire Police. These were then used for training the AI models, as outlined in the Dataset Information section (Data level Information document). Statement on legal, ethical and access issues: The data for the development of the machine learning algorithm were generated using volunteers' fingermarks under full SHU ethical approval ER52762288. Risk of disclosing participants'identities has been significantly mitigated as references to real names and other identifying information have been removed. The fingerprint images were collected under an alphanumeric code. They were exported in JPEG format and shared with the AI team, already anonymised and contained no identifying information about the original donor. The only metadata included with the images was m/z values yielding the particular MALDI mass image of the fingerprint. The model was trained based on the grades assigned by West Yorkshire Police, which are the sole classification information included in the training dataset. Resource language: English Metadata language: English Depositing User: Simona Francese Date Deposited: 24 Apr 2025 09:22 Last Modified: 29 May 2025 14:31