Predicting the Insulin dose Intelligently for perusal by humans- Missing Part-Upload Meal Image for Calories!

#ai #health

Note: This method may have some fuzziness assigned to it, but it’s still not foolproof and needs to be validated. Therefore, it’s not fully reliable. Always seek an opinion on your insulin dose before taking it, or stay within a safe dosage range.

Some of the past works in this area use:

1. Deep Reinforcement Learning
2. Multilinear Regression Models algorithm
3. RNN (LSTM)
4. Artificial Neural Network algorithm
5. Machine Learning
6. Fuzzy Logic

There are many research papers in this area, with reinforcement learning being the latest. But still, when people check their sugar levels, they are not given the correct recommedation insulin dose they should take.

If there is anything you think is apt and correct, comment here and send a link.

So, when we say Machine Learning is used to compute the amount of insulin that can be taken, let’s take it out of research papers.

When it’s needed at people’s homes.

When Fuzzy Logic can provide a safe interval, a ≤ insulin dose ≤ b, for a dose that can be taken.

Sometimes, the insulin dose chosen by the person may be more than necessary, especially when some physical activity is involved.

Or at times the insulin dose taken by the human taker is less, given the meal was heavy.

So how do we combine it all?

Smart watches?

Or

Smart Apps?

Any, but given that blood sugar is measured via mobile apps now, the data goes to the internet, so these computations can be done online to determine the safe dose of insulin for that part of the day. Deciding on the nighttime insulin dose requires extra care and caution.

What all are inputs:

1. Pattern of day, walk, exercise, rest — You can let your smart watch communicate with the glucose app to know it.
2. Food taken: Click the meal image to see its constituents using Gen AI computer vision.
3. Measurements of sugar (already with the App): Already with the online app.
4. Fuzziness: Some fuzziness will remain until the algorithm matures, after which it will be accurate to some degree.

The exact amount can be specified here, and a safe range can be provided so the person can take in that amount, ensuring that the blood sugar level remains around 100 after overnight sleep or above 80 during the day.

Neither hyper nor hypo, for any dose units.

Food taken can be discussed in the app with voice mode on.

Each country can have its own specific meal types.

Or the food type can be clicked in the image, and the mobile app can find out what the meal was- heavy/light or moderate. Deep Learning can even decode the calories from food images. Time to do this? Chat bots?

Gen AI can do it! Try it!

How?

Some training on fitness apps data, and see how it classifies calories as high, low, medium, very low, or very high. Then it can even tell you the exact number of calories.

Once these inputs are provided to the app, it measures the current sugar level and immediately displays the recommended insulin dose in units and the safe interval.

A safe insulin interval is necessary because sometimes the human taker, the patient, might want to make some adjustments. Therefore, following a safe insulin limit with a clear minumum unit and a clear maximum unit is the best practice to ensure safety. Still, we need a nod! A doctor’s nod?

The algorithm can be a deep learning algorithm. Since today’s modern glucose apps are mobile online apps, they have accumulated a lot of data.

However, they have not clicked the food image taken by the patient. This is the missing part.

Let’s add the missing part.

Converting an image to calories and fat isn’t straightforward and isn’t yet available to us with high accuracy. We need to first test this as a model before using it. Validate this part first, then only we can incorporate it into a diabetes app.

It’s not convenient to write out the meal every time, so an image is a better option.

Or it can be semi-supervised — meaning we add pictures and train the model with manual constituents as well. Although this process is cumbersome, it can be bebefactor.

Then the vast amount of data on mobile apps can benefit society.

Only the online meal photographs need to be uploaded.

The app’s settings need to specify the meal type and constituents, carbs, fats, fibres, and more, then automatically determine the calorie intake using gen AI image processing techniques.

Other things, such as a buzzer to remind you that it’s time for a meal or to measure the sugar, are add-ons.

Here it goes, this fills the missing part.

Thank you for reading.

Subscribe for updates.