نبذة مختصرة : U ovom se radu bavimo primjenom metoda dubokog učenja na zadacima procjene zubnog statusa, procjene dobi i spola. Primjena je temeljena na analizi rendgenskih snimaka, na kojima je moguće vidjeti različite faze razvoja zuba, ali i histološke promjene na zubima već završenog razvoja. Korištenjem umjetnih konvolucijskih neuronskih mreža, navedene probleme rješavamo na vrlo efikasan i formalan matematički način. Objašnjen je koncept neuronskih mreža te je dan detaljan opis svih korištenih modela. Godine određujemo metodom regresije, spol metodom klasifikacije, a zubni status korištenjem sustava za lokalizaciju i klasifikaciju objekata YOLO i Faster RCNN. Prosječna greška pri procjeni godina iznosi 4,9 godina, točnost procjene spola je 97,11%, a modeli YOLO i Faster RCNN pri određivanju zubnog statusa ostvaruju mAP od 55,01% i 77,88%, respektivno. ; In this paper we present the application of deep-learning methods in dental status assessment and age and sex estimation. The application of deep-learning methods is based on the analysis of Xray images which show different stages of tooth development, as well as histological changes in fully developed teeth. By using artificial convolutional neural networks, we solve these problems in a very effective and mathematically formal manner. We explain the concept of neural networks and provide a detailed insight into all models used. We use regression to determine age, classification to determine sex, and we assess dental status by utilizing localisation and classification systems such as YOLO and Faster RCNN. The average age estimation error is 4.9 years, and sex estimation accuracy is 97.11%. By using YOLO and Faster RCNN, we obtain the mAP score of 55.01% and 77.88% for dental status assessment, respectively.
Processing Request
No Comments.