Automated Biological Sex Estimation from Two-Dimensional Dental Arch Images Using Deep Learning
This study demonstrates that biological sex can be accurately estimated from radiation-free 2D dental arch images using deep learning, with EfficientNetB0 achieving 96.39% accuracy and highlighting the inter-canine and inter-molar regions as key discriminative features for forensic identification.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are a detective trying to solve a mystery, but the only clue you have left is a set of teeth. In the world of forensics, figuring out if a person was male or female is the first big step to identifying them. Usually, detectives look at the pelvis or the skull, but in disasters like fires or earthquakes, those bones often turn to dust. Teeth, however, are the toughest parts of the human body; they can survive heat and pressure that would destroy everything else. For a long time, the only way to get a good look at teeth for identification was to use X-rays. But X-rays are like a double-edged sword: they give you great detail, but they also expose people to radiation, and you need big, expensive machines to take them. This makes X-rays hard to use in the middle of a disaster zone or in remote villages. So, scientists have been asking: Is there a way to tell the difference between male and female teeth without using any radiation at all?
This paper explores a clever new trick: instead of looking inside the teeth with X-rays, the researchers looked at the shape of the entire row of teeth, known as the dental arch. Think of your teeth not as individual bricks, but as a horseshoe-shaped curve. The study suggests that male arches tend to be wider and more rectangular, while female arches are often narrower and more tapered, almost like a pointed arch. The researchers wanted to see if a computer could learn to spot these subtle shape differences just by looking at a flat, 2D picture of the teeth, without ever needing an X-ray machine.
The team gathered a massive collection of 6,000 digital pictures of dental arches from adults aged 18 to 65. They split these pictures into two groups: one for the computer to learn from, and a "secret test" group it had never seen before. They then trained four different types of artificial intelligence (AI) brains to look at these pictures and guess the sex. These AI brains were like different students with different study habits: one was a simple custom design, one was a lightweight mobile-friendly model, one was a deep, complex model, and one was a highly efficient, modern model called EfficientNet.
The results were surprising and exciting. The most efficient model, EfficientNet, got it right about 96.39% of the time on the secret test. This is just as good as, or even better than, many methods that do use X-rays. The researchers also tested a smaller, lighter model called MobileNetV2, which performed almost exactly the same (95.88% accuracy). This is a big deal because MobileNetV2 is small enough to run on a regular laptop or even a powerful phone, meaning it could be used in the field without needing a hospital's heavy equipment.
The paper also checked how the AI was making its decisions. Using a special visualization tool, they saw that the AI was focusing its attention exactly where human experts say the differences exist: the space between the canine teeth (the "eye teeth") and the space between the back molars. It wasn't just guessing randomly; it was looking at the specific parts of the mouth that are biologically different between men and women.
However, the researchers are careful not to call this a finished, perfect solution just yet. They found that a very deep, complex AI model actually did worse than the simpler ones, likely because it got confused by the lack of detail in the simple pictures. They also noted that their data came from just one place, so the AI might need to be tested on people from different parts of the world to make sure it works everywhere. While the study proves that radiation-free sex estimation is possible and highly accurate, the authors suggest that before this becomes a standard tool for disaster victims, it needs to be tested in real-world field conditions and on more diverse groups of people. But the door is definitely open: it seems we might soon be able to identify people just by the shape of their smile, safely and without any radiation.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.