Pregled bibliografske jedinice broj: 1275782
Image analytics for biological profiling of human skeletal remains: the application of transfer learning-based image analysis using a user-friendly visual programming approach
Image analytics for biological profiling of human skeletal remains: the application of transfer learning-based image analysis using a user-friendly visual programming approach // 12th Annual Meeting of the International Society for Forensic Radiology and Imaging (ISFRI)
Toulouse, Francuska, 2023. str. 116-116 (predavanje, međunarodna recenzija, sažetak, znanstveni)
CROSBI ID: 1275782 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
Image analytics for biological profiling of human skeletal remains: the application of transfer learning-based image analysis using a user-friendly visual programming approach
Autori
Jerković, Ivan ; Jerković, Nika ; Bašić, Željana ; Kružić, Ivana ; Bareša, Tina ; Mladenović, Saša
Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni
Izvornik
12th Annual Meeting of the International Society for Forensic Radiology and Imaging (ISFRI)
/ - , 2023, 116-116
Skup
12th Annual Meeting of the International Society for Forensic Radiology and Imaging (ISFRI)
Mjesto i datum
Toulouse, Francuska, 25.05.2023. - 27.05.2023
Vrsta sudjelovanja
Predavanje
Vrsta recenzije
Međunarodna recenzija
Ključne riječi
image analytics ; transfer learning ; machine learning ; visual programming ; MSCT ; cranium ; sex estimation
Sažetak
Developing biological profiling methods based on morphological trait scores or skeletal measurements is a time-consuming process prone to observers’ inconsistencies and, often, results of limited practical significance. To overcome such issues, artificial intelligence has been introduced to forensic anthropology, but initially without remarkable effect, as these, primarily neural network-based techniques, required large sample sizes and advanced computer science skills. The developments of user-friendly tools have brought these methods to scientists of different profiles, while introducing transfer learning methods has enabled the usage of more realistic sample sizes. In this paper, we present possibilities of applying Orange: Data Mining Toolbox to establish methods that can reconstruct the biological profile of individuals directly from images. In the presented cases, we will use cranial images from documented virtual skeletal collection obtained from medical MSCT scans to demonstrate how to transform images to a vector representation using neural networks previously trained on general image datasets. The data with extracted features will furtherly be used to explore sexual dimorphism using unsupervised machine learning (ML) and construct prototypes of sex classification models by supervised ML methods. Finally, we will show how tools in Orange can help us visualize incorrectly classified specimens and interpret clusters.
Izvorni jezik
Engleski
Znanstvena područja
Računarstvo, Temeljne medicinske znanosti, Sigurnosne i obrambene znanosti, Etnologija i antropologija, Kognitivna znanost (prirodne, tehničke, biomedicina i zdravstvo, društvene i humanističke znanosti)
POVEZANOST RADA
Projekti:
HRZZ-UIP-2020-02-7331 - Forenzička identifikacija ljudskih ostataka analizom MSCT snimaka (CTforID) (Kružić, Ivana, HRZZ - 2020-02) ( CroRIS)
Profili:
Ivana Kružić
(autor)
Željana Bašić
(autor)
Ivan Jerković
(autor)
Saša Mladenović
(autor)
Tina Bareša
(autor)