Mummified pets from Ancient Egypt automatically segmented
Image credit: Dreamstime
Researchers from the University of Malta have developed an algorithm that performs “virtual segmentation” to reveal the internal structures of archaeological remains; specifically, ancient mummified animals.
The hidden internal structures of archaeological remains can be revealed through a technique, Propagation Phase Contrast Synchrotron Microtomography (PPC-SRμCT), which is considered the gold standard for non-invasive and non-destructive imaging.
PPC-SRμCT allows researchers to perform a “virtual autopsy” and “virtual unwrapping” of mummified remains, uncovering information about processes used to create mummies and informing historical study. It has been applied to archaeozoological studies of mummified animal remains from the Ptolemaic and Roman periods of Ancient Egypt, originating from the third century BC to the fourth century AD.
Image segmentation is an important process in computer vision, unlocking higher level tasks such as understanding the progress of a disease from a CT image. Previous studies have trained AI to segment imagery automatically, with most work focused on segmenting medical images, such as CT scans of the human liver.
However, there is a significant difference between conventional CT scans and the volumetric scans captured by PPC-SRμCT. Methods for segmenting the latter are comparatively primitive, requiring an expert to manually segment the virtual specimen to separate different parts and materials.
The University of Malta researchers developed a tool to automatically segment volumetric images, aiming to drastically reduce the effort involved from weeks of effort, even for small remains, to a matter of hours. They used manually segmented samples from previous work to train and tune their machine learning model.
For a set of four specimens of Ancient Egyptian animal mummies, they achieved an overall accuracy of 94 to 98 per cent when compared with manually segmented slices, approaching the off-the-shelf commercial software harnessing deep learning (97 to 99 per cent) at much lower complexity.
A qualitative analysis of the segmented output demonstrated that their results were close in terms of usability to those from deep learning, justifying the use of these techniques. Their system has the key advantage of achieving high accuracy at lower complexity, allowing it to scale well for larger volumes.
The researchers wrote in their study that their automatic segmentation algorithm could be applied directly to a variety of other applications in which microtomography is used.
“In industrial applications automatic segmentation will be useful in non-destructive metrology and detection of voids, cracks and defects,” they wrote. “In biomedical research, it would be useful in small animal and tumour imaging, amongst others. Other applications include nanotechnology, geology, electronics and food science.”
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.