Artificial Intelligence in Forensic Science: A Systematic Review and Meta-Analysis of Opportunities and Limitations
- 1Forensic Medicine and Clinical Toxicology Department, Faculty of Medicine, Zagazig University, Egypt
- 2Forensic Medicine and Clinical Toxicology Department, Faculty of Medicine, Zagazig University
Abstract
Background: Forensic science is increasingly incorporating artificial intelligence (AI) to improve analytical precision and support decision-making. However, comprehensive quantitative synthesis remains limited, and reported performance varies across forensic disciplines. Objectives: To systematically examine and quantitatively synthesize evidence on AI applications in crime scene reconstruction, firearm injury analysis, and forensic radiology, focusing on algorithmic techniques, sources of heterogeneity, and diagnostic performance. Methods: We systematically searched PubMed and ScienceDirect for studies published from 2010 to 2025. Eligible studies applied AI-based methods in forensic contexts and reported quantitative performance outcomes. Study selection followed PRISMA 2020 guidelines. Qualitative and quantitative approaches were used for data synthesis. Random-effects meta-analyses were performed for comparable outcomes, including accuracy and odds ratios (ORs). Heterogeneity was assessed using Cochran’s Q and I² statistics. Results: AI demonstrated high performance across several forensic applications. In crime scene reconstruction, pooled analysis of two studies showed an overall accuracy of 96.2% (95% CI: 0.86–0.99; P < 0.001). In firearm injury analysis, the pooled OR for discriminating between entrance and exit wounds was 2.03 (95% CI: 0.66–6.26; P = 0.22). In forensic radiology, pooled analysis of sex determination was not statistically significant (OR = 1.63; 95% CI: 0.95–2.99; P = 0.159) and showed substantial heterogeneity (I² = 75.8%). In contrast, pooled accuracy for sex classification across ten studies was 91.0% (95% CI: 0.89–0.95; P < 0.001), with moderate heterogeneity (I² = 55.0%). Deep learning generally outperformed traditional machine learning, particularly in image-based applications such as forensic radiology. Conclusions: AI demonstrates promising performance across several forensic disci Background: Forensic science is increasingly incorporating artificial intelligence (AI) to improve analytical precision and support decision-making. However, comprehensive quantitative synthesis remains limited, and reported performance varies across forensic disciplines. Objectives: To systematically examine and quantitatively synthesize evidence on AI applications in crime scene reconstruction, firearm injury analysis, and forensic radiology, focusing on algorithmic techniques, sources of heterogeneity, and diagnostic performance. Methods: We systematically searched PubMed and ScienceDirect for studies published from 2010 to 2025. Eligible studies applied AI-based methods in forensic contexts and reported quantitative performance outcomes. Study selection followed PRISMA 2020 guidelines. Qualitative and quantitative approaches were used for data synthesis. Random-effects meta-analyses were performed for comparable outcomes, including accuracy and odds ratios (ORs). Heterogeneity was assessed using Cochran’s Q and I² statistics. Results: AI demonstrated high performance across several forensic applications. In crime scene reconstruction, pooled analysis of two studies showed an overall accuracy of 96.2% (95% CI: 0.86–0.99; P < 0.001). In firearm injury analysis, the pooled OR for discriminating between entrance and exit wounds was 2.03 (95% CI: 0.66–6.26; P = 0.22). In forensic radiology, pooled analysis of sex determination was not statistically significant (OR = 1.63; 95% CI: 0.95–2.99; P = 0.159) and showed substantial heterogeneity (I² = 75.8%). In contrast, pooled accuracy for sex classification across ten studies was 91.0% (95% CI: 0.89–0.95; P < 0.001), with moderate heterogeneity (I² = 55.0%). Deep learning generally outperformed traditional machine learning, particularly in image-based applications such as forensic radiology. Conclusions: AI demonstrates promising performance across several forensic disciplines, particularly in crime scene reconstruction and forensic radiology classification tasks. However, heterogeneity and limited generalizability remain challenges due to variations in study design, outcome reporting, and validation techniques. Standardized methodologies and external validation are needed to support reliable integration of AI into forensic practice.
Related
- Bridging the Translational Gap in Artificial Intelligence for Otology and Neurotology: Clinical Readiness, Limitations, and Research Priorities
- Early insights into nurses' perceptions, opportunities, and concerns regarding Ambient artificial intelligence in nursing practice
- Circulating Biomarkers with Artificial Intelligence for Clinical Risk Assessment and Therapeutic Monitoring: Evidence, Limitations, and Translation
- Current Advances in Spatial Pathology of Nasal Polyps: Challenges, Opportunities for Artificial Intelligence, and Future Directions
- Artificial Intelligence as a Catalyst for Novel Therapeutic Opportunities in Precision Oncology.
BibTeX
@article{elkhateeb2026artificial,
title = {{Artificial Intelligence in Forensic Science: A Systematic Review and Meta-Analysis of Opportunities and Limitations}},
author = {Shereen Elkhateeb and Doaa Abdelrazzak and Nema allah Hosieny},
journal = {Zagazig Journal of Forensic Medicine and Toxicology},
year = {2026},
month = oct,
doi = {10.21608/zjfm.2026.485051.1236},
url = {https://doi.org/10.21608/zjfm.2026.485051.1236},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial Intelligence in Forensic Science: A Systematic Review and Meta-Analysis of Opportunities and Limitations
AU - Shereen Elkhateeb
AU - Doaa Abdelrazzak
AU - Nema allah Hosieny
PY - 2026
DA - 2026/10/25
JO - Zagazig Journal of Forensic Medicine and Toxicology
DO - 10.21608/zjfm.2026.485051.1236
UR - https://doi.org/10.21608/zjfm.2026.485051.1236
AB - Background: Forensic science is increasingly incorporating artificial intelligence (AI) to improve analytical precision and support decision-making. However, comprehensive quantitative synthesis remains limited, and reported performance varies across forensic disciplines. Objectives: To systematically examine and quantitatively synthesize evidence on AI applications in crime scene reconstruction, firearm injury analysis, and forensic radiology, focusing on algorithmic techniques, sources of heterogeneity, and diagnostic performance. Methods: We systematically searched PubMed and ScienceDirect for studies published from 2010 to 2025. Eligible studies applied AI-based methods in forensic contexts and reported quantitative performance outcomes. Study selection followed PRISMA 2020 guidelines. Qualitative and quantitative approaches were used for data synthesis. Random-effects meta-analyses were performed for comparable outcomes, including accuracy and odds ratios (ORs). Heterogeneity was assessed using Cochran’s Q and I² statistics. Results: AI demonstrated high performance across several forensic applications. In crime scene reconstruction, pooled analysis of two studies showed an overall accuracy of 96.2% (95% CI: 0.86–0.99; P < 0.001). In firearm injury analysis, the pooled OR for discriminating between entrance and exit wounds was 2.03 (95% CI: 0.66–6.26; P = 0.22). In forensic radiology, pooled analysis of sex determination was not statistically significant (OR = 1.63; 95% CI: 0.95–2.99; P = 0.159) and showed substantial heterogeneity (I² = 75.8%). In contrast, pooled accuracy for sex classification across ten studies was 91.0% (95% CI: 0.89–0.95; P < 0.001), with moderate heterogeneity (I² = 55.0%). Deep learning generally outperformed traditional machine learning, particularly in image-based applications such as forensic radiology. Conclusions: AI demonstrates promising performance across several forensic disci Background: Forensic science is increasingly incorporating artificial intelligence (AI) to improve analytical precision and support decision-making. However, comprehensive quantitative synthesis remains limited, and reported performance varies across forensic disciplines. Objectives: To systematically examine and quantitatively synthesize evidence on AI applications in crime scene reconstruction, firearm injury analysis, and forensic radiology, focusing on algorithmic techniques, sources of heterogeneity, and diagnostic performance. Methods: We systematically searched PubMed and ScienceDirect for studies published from 2010 to 2025. Eligible studies applied AI-based methods in forensic contexts and reported quantitative performance outcomes. Study selection followed PRISMA 2020 guidelines. Qualitative and quantitative approaches were used for data synthesis. Random-effects meta-analyses were performed for comparable outcomes, including accuracy and odds ratios (ORs). Heterogeneity was assessed using Cochran’s Q and I² statistics. Results: AI demonstrated high performance across several forensic applications. In crime scene reconstruction, pooled analysis of two studies showed an overall accuracy of 96.2% (95% CI: 0.86–0.99; P < 0.001). In firearm injury analysis, the pooled OR for discriminating between entrance and exit wounds was 2.03 (95% CI: 0.66–6.26; P = 0.22). In forensic radiology, pooled analysis of sex determination was not statistically significant (OR = 1.63; 95% CI: 0.95–2.99; P = 0.159) and showed substantial heterogeneity (I² = 75.8%). In contrast, pooled accuracy for sex classification across ten studies was 91.0% (95% CI: 0.89–0.95; P < 0.001), with moderate heterogeneity (I² = 55.0%). Deep learning generally outperformed traditional machine learning, particularly in image-based applications such as forensic radiology. Conclusions: AI demonstrates promising performance across several forensic disciplines, particularly in crime scene reconstruction and forensic radiology classification tasks. However, heterogeneity and limited generalizability remain challenges due to variations in study design, outcome reporting, and validation techniques. Standardized methodologies and external validation are needed to support reliable integration of AI into forensic practice.
SN - 2536-9849
SN - 1687-160X
ER -