The AI Mole Mapper

Here's an idea that came to me recently as I was thinking of novel use cases for single-board computers and benefitting society.  It came shortly after another idea describing how to use the vast range of low-cost sensors available today to help identify health risks over the long term.

Introduction:

Mole mapping is a technique used to raise an alarm when there are changes to a person’s skin features. Various services exist just search for "mole mapping" online and you will get the idea, where digital photographs are taken of the whole body then compared to historic images perhaps annually or every few years to identify any changes that should be investigated further. Such services require the patient to spend time at the studio, naked and at financial expense. Wouldn't it be better if such a service could be conducted in the privacy of your own home on a daily basis to catch signs of cancer at the earliest possible opportunity? Rather than monitoring only one person, it could also keep a check on anyone else in the household.

The idea:

Using a single board computer fitted with an industrial grade camera (not the types hobbyists use with image processors built in which remove details humans typically ignore) which takes highly detailed raw quality images. The device would be placed in the bathroom, probably running from the shaver power socket, no Wi-Fi connectivity for privacy compliance to alleviate any fears of it constantly 'recording' images of occupants. The bathroom is probably the ideal location as it's where folk are most likely not to be wearing clothes. The device would recognize occupants using a simple AI algorithm and then record a map of their skin (not a photograph for privacy reasons) using the amazing features of AI or regular CV type processing.  It doesn't require vast high-performance processors as it can crunch the data in its own time from a cache slower than real-time high fps frame rates. The output of the photo analysis which will be retained internally to the device may simply be the X/Y coordinates and shade of any skin features. The next time that person steps in front of the device it creates a similar map and compares it against historic maps. If it detects new features on the subject's body it could simply beep, flash a light or direct the user by audio instructions to point up/down/left/right to the area of concern. It may simply be a new tattoo, or a pen leaked in your top pocket or you bruised your arm playing sport and the device can be instructed to ignore it or the AI would do this automatically depending on its training but, what if it’s something that should be looked at in more detail?! Then off to the doctor to have that particular area checked out.

Unfortunately, with an ever-increasing risk of cancers, people still spend too much time in the sun and may not pay enough attention to their bodies etc. this device will hopefully detect early signs of skin changes, lumps, and bumps which is known to improve the outcome if diagnosed early. Such a simple product with massive benefits.

The type of hardware I had in mind:

It would make use of the AI neural processor which is common to many boards these days to recognize who is in the room, it wouldn't necessarily need to be the top-of-the-range NPU (Naural Processing Unit) either as its main job is to recognize a person in a quiet environment. It would utilize the RTC functionality to let the board sleep and consume minimal power or a PIR to the same interrupt line, so it is awoken when there is a presence. It makes use of the camera interface and industrial-style cameras without onboard ISP, perhaps low light cameras and global shutter to avoid distorted images and assist the software processing as much as possible at the data source. It may use the encryption engine to store any info securely, although not particularly required for the reason that initially comes to mind as naked images aren't being kept once processed from RAM as it is all obfuscated to X/Y coordinates however the map output could potentially be used for personal identification in a similar way as dental records thus ought to be encrypted to be GDPR compliant.

Maybe, someone, will one day create such a gadget, it could save lives and isn't particularly difficult to build.