Pelvic classification based on deep learning algorithm on clinical CT scans in Malaysian population
The estimation of biological sex and skeletal age is vital when dealing with skeletal remains. As human are sexually dimorphic, are present in the skeleton, markedly after the age of puberty. Age related changes also can be quantified in the skeleton, manifesting in the formation of skeleton to a...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/60222/1/YASMIN%20ARIJAH%20BINTI%20CHE%20YAHAYA-E.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The estimation of biological sex and skeletal age is vital when dealing with skeletal remains.
As human are sexually dimorphic, are present in the skeleton, markedly after the age of puberty.
Age related changes also can be quantified in the skeleton, manifesting in the formation of
skeleton to adulthood. Pelvis bone is the most trustworthy part in human body for sex
estimation and age classification. In this research, Phenice method will be utilised for the sex
estimation and age classification. The utility of deep convolutional neural network (DCNN)
for sex and age estimation was evaluated using images generated from reconstructed 3-
dimensional computed tomography images. This study analysed the Phenice method by
utilising 3D CT scans by deep learning algorithm for sex estimation and age estimation.
The CT scans of 290 individuals (179 males and 111 females) which comprised an age range
from 7 to 94 years old of the Malaysian population were analysed by GTM (Google Teachable
Machine). The sample was collected at Hospital Universiti Sains Malaysia (HUSM) starting
from 2009 until May 2023. The 2D images screenshots of CT scans were reconstructed to 3D
model using Invesalius 3.1 and PicPick for captured images for learning and testing. The
samples have been separated into four features, which are the ventral arc, the subpubic
concavity, the medial aspect of ischiopubic ramus and overall features of Phenice method. For
age classification, each feature has been divided into two main groups which are age above 20
years old and age below 20 years old.
The Phenice sex estimation method provides 98% of mean precision while 88.3% and 95% for
mean sensitivity and mean specificity respectively. However, the Phenice age classification
method is only applicable for sample age above 20 years old. It gives 97.75% of mean precision, 93.95% of mean sensitivity and 95.7% of mean specificity. For samples under 20
years old, the precision, sensitivity and specificity cannot be calculated as the result by Google
Teachable Machine is error.
This research concludes that using Google Teachable Machine provide high accuracy and
precision for sex estimation but not useful for age classification for sample below 20 years old. |
|---|