Publication: Lumbar Vertebraların Bilgisayarlı Tomografi Görüntülerinden Ölçülen Yükseklik ve Derinlik Değerlerinin Yapay Zeka İle Cinsiyet Tayininde Kullanılabilirliği
Abstract
Cinsiyet tayini, kimlik tespitinde en önemli adımlardan biridir. Cinsiyet tayininde, güvenirliliği kanıtlanmış olan pelvis ve kafatası kemikleri sıklıkla kullanılır. Fakat bazen bu kemiklere olay yerinde ulaşılamayabilir. Bu gibi durumlarda vertebralar, sayıca çok olan ve daha iyi korunabilen kemikler olarak cinsiyet tahmininde kullanılmıştır. Bu çalışma ile, bilgisayarlı tomografiden elde edilen lumbar vertebra görüntüleri üzerinde işaret noktaları arasındaki uzunluk ve açı değerlerini hesaplayarak, makine öğrenme sınıflandırıcıları ile cinsiyet tahmini yapmayı amaçladık. Ondokuz Mayıs Üniversitesi Radyoloji Ana Bilim Dalı arşivinden alınan, hiçbir anomalisi olmayan 20-40 yaş aralığında 50 kadın 50 erkek hastanın lumbar vertebra bilgisayarlı tomografi görüntüleri kullanıldı. Digital Imaging and Communications in Medicine (DİCOM) formatındaki görüntüler kişisel iş istasyonunda (Horos Project, Version 3.0) ortogonal düzleme getirildi. Makine öğrenmesi algoritmalarını kullanmak amacıyla geliştirilmiş olan Sekazu adlı programa yüklenen görüntüler üzerinde anatomik noktalar elle etiketlenerek uzunluk ve açı değerleri hesaplandı. 13 farklı makine öğrenme sınıflandırıcısı kullanılarak, doğruluk oranları hesaplandı. Bu hesaplamaların sonucuna göre L1-L5 vertebralarının cinsiyet tayininde doğruluk oranları sırasıyla %93, %90, %90, %87 ve %86 olarak bulundu. Bu veriler, K-Nearest Neighbors, Extra Tree, Random Forest, Decision Tree, Gaussian Bayes ve Linear Discrimant Classifier makine öğrenme sınıflandırıcıları ile bulundu. Elde edilen sonuçlara göre lumbar vertebra üzerinde etiketlenen noktalar ve makine öğrenmesi uygulaması sonucunda %86-%93 aralığında doğruluk oranları ile cinsiyet tahmini yapılabileceği bulundu. Cinsiyet tahmininde doğruluk oranı en yüksek vertebranın L1 olduğu sonucuna ulaşıldı. L1-L5 omurlarının herbiri için yapılan ölçümler sonucunda, parametrelerin bir çoğunda erkeklerden elde edilen ortalama değerlerin kadınlardan daha büyük ve istatistiksel olarak anlamlı olduğu tespit edilerek bel omurlarının cinsel dimorfik özellikte olduğu tespit edildi.
Gender determination is one of the most important steps in identification. Pelvis and skull bones, which have proven reliability, are frequently used in sex determination. However, sometimes these bones cannot be reached at the crime scene. In such cases, vertebrae have been used for sex estimation as bones that are more numerous and better preserved. In this study, we aimed to predict gender using machine learning classifiers by calculating the length and angle values between landmarks on lumbar vertebra images obtained from computed tomography. Lumbar vertebra computed tomography images of 50 female and 50 male patients aged 20-40 years who had no anomaly and were taken from the archive of Ondokuz Mayıs University Radiology Department were used. Images in Digital Imaging and Communications in Medicine (DICOM) format were brought to the orthogonal plane on the personal workstation (Horos Project, Version 3.0). The length and angle values were calculated by manually labeling the anatomical points on the images uploaded to the program called Sekazu, which was developed to use machine learning algorithms. Accuracy rates were calculated using 13 different machine learning classifiers. According to the results of these calculations, the accuracy rates in sex determination of L1-L5 vertebrae were found to be 93%, 90%, 90%, 87% and 86%, respectively. These data were found with K-Nearest Neighbors, Extra Tree, Random Forest, Decision Tree, Gaussian Bayes and Linear Discrimant Classifier machine learning classifiers. According to the results obtained, it was found that the tagged points on the lumbar vertebrae and the machine learning application could make gender prediction with an accuracy rate of 86%-93%. It was concluded that the vertebra with the highest accuracy rate in estimating gender was L1. As a result of the measurements made for each of the L1-L5 vertebrae, it was determined that the mean values obtained from men in most of the parameters were larger and statistically significant than women, and it was determined that the lumbar vertebrae were sexually dimorphic.
Gender determination is one of the most important steps in identification. Pelvis and skull bones, which have proven reliability, are frequently used in sex determination. However, sometimes these bones cannot be reached at the crime scene. In such cases, vertebrae have been used for sex estimation as bones that are more numerous and better preserved. In this study, we aimed to predict gender using machine learning classifiers by calculating the length and angle values between landmarks on lumbar vertebra images obtained from computed tomography. Lumbar vertebra computed tomography images of 50 female and 50 male patients aged 20-40 years who had no anomaly and were taken from the archive of Ondokuz Mayıs University Radiology Department were used. Images in Digital Imaging and Communications in Medicine (DICOM) format were brought to the orthogonal plane on the personal workstation (Horos Project, Version 3.0). The length and angle values were calculated by manually labeling the anatomical points on the images uploaded to the program called Sekazu, which was developed to use machine learning algorithms. Accuracy rates were calculated using 13 different machine learning classifiers. According to the results of these calculations, the accuracy rates in sex determination of L1-L5 vertebrae were found to be 93%, 90%, 90%, 87% and 86%, respectively. These data were found with K-Nearest Neighbors, Extra Tree, Random Forest, Decision Tree, Gaussian Bayes and Linear Discrimant Classifier machine learning classifiers. According to the results obtained, it was found that the tagged points on the lumbar vertebrae and the machine learning application could make gender prediction with an accuracy rate of 86%-93%. It was concluded that the vertebra with the highest accuracy rate in estimating gender was L1. As a result of the measurements made for each of the L1-L5 vertebrae, it was determined that the mean values obtained from men in most of the parameters were larger and statistically significant than women, and it was determined that the lumbar vertebrae were sexually dimorphic.
Description
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
124
