What’s the calculation of automatic flushing, warm air drying, and seat insulation? Researchers at Stanford University in the United States reported a set of software and hardware that can track the health and disease indicators of users’ excreta, which can be called a truly intelligent toilet.

According to the paper published in Nature Biomedical Engineering on April 6, this set of conceptual smart devices can be retrofitted into existing On the ordinary toilet, it is constructed and tested for male subjects, including three modules: urine analysis, urine flow analysis and fecal analysis.

One of the novelties of the research is to cooperate with the camera and train artificial intelligence vision to analyze urine flow and fecal morphology, and use “anal pattern recognition” to supplement the embedded Fingerprint recognition on the water button.

Yann LeCun, one of the three giants of deep learning and winner of the Turing Awards in 2019, commented on social media: “Convolutional network toilets are veritable. To benefit mankind. “


Automatic retractable test strip

Human excreta is an important health reflection, but it is difficult to achieve long-term and sustained clinical monitor. The Stanford team chose to start with the toilet and perform non-invasive tests on urine and feces.

From the schematic view, the smart toilet is equipped with 7 hardware, including 1 pressure sensor, 1 motion sensor, 1 set of urine analysis test strips, and 4 cameras .



Stanford Smart Toilet. (I) Pressure sensor (ii) Motion sensor (iii) Urine test strip (iv) Fecal camera (v) Anal camera (vi) Urine flow camera

Among them, the test strip can be used for qualitative and semi-quantitative analysis of 10 biomarkers, including red blood cells, post-urinary bilirubinogen, bilirubin, protein, sub Nitrate, ketone, glucose, pH, urine specific gravity and white blood cells.

Suppose there is an adult male aiming to urinate in the center of the toilet, trigger the infrared motion sensor, the test strip will automatically extend. About 30-60 seconds, the test strip is saturated with urine and then automatically shrinks back to its original position. A camera will capture video at the same time, perform real-time motion analysis, and then wirelessly transmit the data to the cloud storage.

After urination, the test strip will automatically fall off and be discarded in the toilet. Taking environmental protection into consideration, the test strips replaced the plastic materials with water-soluble polysaccharide materials.


Camera captures “Tick and Tick”

For urinary flow analysis, the researchers wanted to find a way beyond the traditional flowmeter and use computer vision For power, two wide-angle cameras were installed, 960 × 1280 pixels, 240 frames per second.

In order to set the parameters, the researchers invited 10 men between 19 and 40 years old to perform 31 urination tests, each with a flow rate ranging from 50 ml to 670 ml.

The results show that the urine flow and urination time estimated by the computer based on the video are highly correlated with the standard flowmeter, and can capture the end of the urination tail that the flowmeter cannot Intermittent droplets. Researchers believe this can provide additional information about the function of the prostate and bladder.


Artificial intelligence to distinguish stool

The shape and hardness of feces can be used for clinical diagnosis of cancer, but there is no established standard. To this end, the researchers invited two enteric surgeons to independently mark different stool images, and the results showed consistency.

Using these images, scientists have taught artificial intelligence 10 classes of “discriminating stools” to teach normal stools, too hard stools (prompt for constipation and other symptoms) and too soft / Liquid feces (prompting for symptoms such as diarrhea), AUC (an indicator to judge the authenticity of the detection method in the test, the closer to 1 the better) is above 0.89.

In addition, the toilet is equipped with a pressure sensor, which is used to calculate the start time of defecation and trigger the camera to collect images. Subsequently, the convolutional neural network can determine the time interval from the beginning to the first bowel movement, which is closely related to the overall function of the intestine. The end time of defecation is determined by toilet paper use or standing up.

In the test, a total of 5 women and 6 men participated in 55 bowel movements.


User identification

Finally, there is another key issue.

Several people may share a toilet, so how to create a personalized health file for users? The researchers embedded a set of fingerprint recognition in the flush button. When the system determines that the toilet is over, the green button lights up on the flush button.

However, in the actual situation, there is no guarantee that the person who presses the flush button must be a person who uses the toilet. Some toilets on the market also have an automatic flush function.

Therefore, the Stanford Smart Toilet also has a unique “anal pattern recognition” function, which uses each person’s anus image to safely match the collected health data.

The author said that the potential health benefits of this toilet system need to be evaluated through large-scale clinical studies, and the system itself needs to be optimized based on the baseline data of human excreta .

As far as key privacy issues are concerned, the paper states that the data will be safely stored in the cloud system and sent to healthcare workers safely.

The research team is also deeply aware that the prospect of this smart toilet is closely related to user acceptance. In a survey of 300 people in the Stanford community (of course, there is a serious deviation in education level), 15.33% said “very acceptable”, 37.33% said “acceptable”, and 30.00% said “uncomfortable”. Among them, the most popular module is “urine analysis” without camera participation, and the least popular module is “anal pattern recognition”.