Deep learning-based analysis of dental X-ray images for forensic estimation of age and sex

doctoral thesis

• Title (croatian): Analiza stomatoloških rendgenskih snimaka za forenzičku estimaciju dobi i spola temeljena na dubokom učenju
• Author: Denis Milošević
• Mentor: Marko Subašić (mentor)
• Committee member: Sven Lončarić (predsjednik povjerenstva)
• Committee member: Zvonko Kostanjčar (član povjerenstva)
• Committee member: Marin Vodanović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Electronic Systems and Information Processing); Zagreb
• Defense date and country: 2022-11-11, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing; Data Processing
• Universal decimal classification (UDC): 004 - Computer science and technology. Computing. Data processing

Abstract

Dental remains are one of the most resistant elements to blunt force, fire, bacterial decomposition, and other external factors, making them a prime candidate for forensic analysis. The forensic analysis aims to identify human remains, with the estimation of age and assessment of sex at the forefront. Current methods of forensic odontology are based on time-intensive, precise manual measurements paired with reference tables and atlases, which can lead to human error. Artificial intelligence and computer vision research shows unprecedented success in many domains, including medical image analysis. This research explores the intersection of deep learning-based methods for computer vision and forensic odontology. Two neural network architectures are proposed, one based on state-of-the-art architectures used as feature extractors paired with attention, and one fully built around the BiFPN module. The hyperparameter space is exhaustively examined, and the best models are compared to the current state of literature. Models are developed for age and sex estimation, tooth detection and type determination, and segmentation of tooth and status. Imperfect teeth with alterations are not excluded, and their impact is extensively examined. The proposed methods perform as well or better than current methods in literature while being fully automated, reproducible, and multiple orders of magnitude faster. The contributions of the thesis are deep learning-based models for sex assessment (classification), age estimation (regression), and region-proposal for tooth status assessment from dental x-ray images.

Abstract (croatian)

Zubni ostaci su jedni od najstabilnijih elemenata u odnosu na silu, vatru, bakterijsku razgradnju i druge čimbenike, što ih čini odličnima za forenzičku analizu. Forenzičkom analizom se žele identificirati ljudski ostaci, pri čemu se ističe određivanje dobi i spola. Trenutne metode forenzičke odontologije se temelje na dugotrajnim metodama ručnog mjerenja i usporedbama s referentnim tablicama i atlasima, što često uvodi ljudske pogreške. Istraživanja na području umjetne inteligencije i analize slike pokazuju odlične rezultate u mnogim područjima, uključujući analizu medicinskih slika. Provedeno istraživanje je interdisciplinarni pristup između forenzičke odontologije i računarske znanosti. Predložene su dvije arhitekture neuronskih mreža. Prva spaja mehanizam pažnje s najuspješnijim arhitekturama analize slika za određivanje značajki, a druga je u izgrađena oko BiFPN modula. Hiperparametarski prostor je iscrpno analiziran, i najuspješniji modeli su uspoređeni s najboljim metodama u literaturi. Razvijeni su modeli za procjenu dobi i spola, za detekciju zuba i određivanje tipa, te za segmentaciju zubi i zubnog statusa. Nesavršeni zubi nisu isključeni iz ovog istraživanja, te je njihov utjecaj analiziran. Predložene metode su automatizirane, ponovljive i značajno brže, te postižu jednake ili bolje rezultate kao najbolje metode iz literature. Doprinosi ove disertacije su modeli temeljeni na dubokim konvolucijskim neuronskim mrežama za procjenu spola (klasifikacija), dobi (regresija), te za prijedlog regija za dentalni status iz rendgenskih snimaka zuba.

• Keywords: Computer vision; Artificial Intelligence; Neural network; Forensic odontology; Age estimation; Sex assessment; Tooth type determination; Teeth detection; Segmentation of teeth and dental alterations
• Keywords (croatian): Umjetna inteligencija; Neuronske mreže; Forenzička odontologija; Procjena dobi; Procjena spola; Određivanje tipa zuba; Detekcija zuba; Segmentacija zuba i dentalnih intervencija
• Language: english
• URN:NBN: urn:nbn:hr:168:663883
• Promotion: 2022
• Study programme: Title: Computer Science; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti (Doktor znanosti)
• Type of resource: Text
• Extent: x, 153 str. : graf. prikazi ; 30 cm. + CD-ROM.
• File origin: Born digital
• Access conditions: Open access
• Created on: 2022-12-14 09:56:01