Phenotuple

The app answers questions about the health problems that an herbal recipe is used for. Specifically, the user selects the herbs which make up his herbal recipe. The relative weights of the composing herbs can be important for the recipe's indications. So, next, the user sets the weights of the chosen herbs. The default setting for the weight profile is flat, meaning that if users do not bother to change weights, then the herbs in the recipe are equally weighted. Finally, the user makes the query, and the Phenotuple server returns the conditions the recipe is used for.

The outputs of Phenotuple are in the form of graphics as well as text. In the case of graphics, a circle - a wheel with 18 spokes - is presented. Each spoke represents a category of diseases. For example, cancerous diseases are a category and cardiovascular diseases belong to another category, and so on. A shortened spoke will make the outer rim of the wheel non-circular. Imagine a flattened tire. Making use of this visual property, when we see a not round circle with shortened spokes, we know the categories of the health problems that the recipe is used to treat for. In the case of text, the top 12 diseases that the recipe is used for are listed.

For advanced use of the app, the user wants to compare differences between two herbal recipes. For example, recipe #1 contains A, B, C; and recipe #2 contains only A, B. Or, in another scenario, the two recipes may contain exactly the same types of herbs but their weight ratios differ. Phenotuple facilitates such comparison by the use of overlapping polygons of different color shades.

All the herbs in the app are in the form of concentrated herbal extract granules, so are the herbal formulas. This form of herbs and formulas are available in many regions and countries of the world. The total number of different herbs in the app is 415 and that of formulas is 303. In addition, acupuncture is also included so that your recipe can include herbs, formulas, as well as acupuncture. The collection of herbal prescriptions and the corresponding disease diagnoses are from the national health insurance reimbursement database (Taiwan) during the decade from 2004 to 2013. The total number of herbal prescription and disease diagnosis pairs is 340 millions.

The algorithm behind Phenotuple is artificial intelligence (AI). Specifically, the herbal prescriptions are converted into images and the corresponding disease diagnoses into labels. The 340 million pairs of images and labels are then trained using state of the art AI algorithm, namely, deep residual convolutional neural network classifiers. The developed AI neural network classifies not only diseases, but also sex, age of the patient and season of the prescription. It is technically called a multitask classifier.