Banca de DEFESA: Renato Queiroz Nogueira Lira

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : Renato Queiroz Nogueira Lira
DATE: 13/12/2024
TIME: 14:00
LOCAL: Plataforma Teams
TITLE:

Application of Deep Learning in the Classification of Wounds Caused by Firearm Projectile

KEY WORDS:

Forensic Medicine, Machine Learning, Gunshot Wounds.

PAGES: 100
BIG AREA: Ciências da Saúde
AREA: Odontologia
SUMMARY:

“In Brazil and around the world, the use of firearms represents the leading cause of violent deaths. Therefore, to investigate these deaths, it is crucial to understand the entire dynamic related to how the crime occurred. However, in many cases, forensic teams, both at the crime scene and during the autopsy, are unable to recover elements sufficient for more robust conclusions, leaving only the traces on the victim to be observed and analyzed. Furthermore, to correctly classify the criminal offense, it is necessary for the analyses performed by the experts to indicate the circumstances under which the shots were fired, such as which wounds correspond to entries and exits and their distances from the victims. With the aim of contributing to these analyses, a study was developed, resulting in an article. In this article, 59 neural convolutional networks for image classification were trained to differentiate between entry and exit wounds and determine the shooting distances through photographs and documents of fatal victim cases examined by the expert teams of the Civil Police of the Federal District between 2012 and 2022. A comprehensive database was constructed with 2,551 images, including 1,883 entry wounds and 668 exit wounds. The ResNet152 architecture demonstrated superior performance in both entry and exit wound classification and medico-legal shooting distance categorization. For the first, it achieved accuracy, recall, F1-score, and specificity of up to 86.90% and an AUC of 82.09%. For the medico-legal shooting distance classification, the ResNet152 showed an accuracy of up to 92.48%, although sample imbalance affected other metrics such as recall and F1- score. Our findings highlight the challenges of standardizing wound images due to varying capture conditions but reflect the practical realities of forensic work. This research underscores the significant potential of deep learning in enhancing forensic medicine practices, advocating for artificial intelligence as a supportive tool to complement human expertise in forensic investigations”

COMMITTEE MEMBERS:
Externo à Instituição - CASIMIRO ABREU POSSANTE DE ALMEIDA - UFRJ
Externo à Instituição - RHONAN FERREIRA DA SILVA - UFC
Presidente - 3437282 - ANDRE FERREIRA LEITE
Externo à Instituição - FÁBIO WILDSON GURGEL COSTA - UFC
Externo ao Programa - 3373847 - MALTHUS FONSECA GALVAO - null