Doctors often look at blood cells under a microscope to diagnose diseases like acute myeloid leukemia, or AML.

AML is a type of blood cancer that affects how white blood cells grow and behave. Finding it early can save lives. But checking blood cells by hand takes time and years of training.

Imagine a doctor looking at hundreds of tiny cell images every day, trying to spot small differences. It is a bit like trying to find a few bad apples in a huge pile of good ones. Scientists are now asking if computers can help with this task.

A recent research from Rutgers and Columbia University study explored a new idea called quantum machine learning. The goal was to see if this technology could help tell the difference between healthy blood cells and leukemia cells using microscope images .

How computers learn from examples

Most people use computers that follow clear instructions. Machine learning works differently. Instead of being told exactly what to do, the computer learns by looking at many examples.

Think of how you learn to recognize handwriting. You do not memorize one letter shape. You see many versions and slowly learn what an “A” usually looks like. Machine learning works in a similar way. It looks at many images of blood cells and learns common patterns.

Today, doctors and researchers mostly use normal computers for this job. These computers are very good, but they often need thousands of examples and a lot of computing power.

What makes quantum computers special

Quantum computers work in a strange but interesting way. Normal computers use bits that are either 0 or 1. Quantum computers use qubits, which can act like 0 and 1 at the same time.

A simple example is flipping a coin. A normal computer waits until the coin lands as heads or tails. A quantum computer works as if the coin is spinning in the air and is both at once. This allows quantum computers to explore many possible answers at the same time.

However, quantum computers are very delicate. If you check them too often, they lose their special behavior. Because of this, many normal learning methods do not work well with them.

Why scientists tested new learning methods

Most machine learning systems learn using a method called backpropagation. You can think of it like correcting homework. The computer checks its answers, sees what it got wrong, and adjusts itself step by step.

This process works great on normal computers, but it does not fit well with quantum systems. Measuring a quantum system too much is like stopping the spinning coin too early.

So scientists designed learning methods that avoid this problem. This study tested two such approaches on real medical images.

The images and data

The researchers used a large public collection of blood cell images from hospitals. Experts had already labeled each image as either healthy or AML. The full dataset included more than 18,000 images from 200 patients.

To make the test realistic, the researchers used smaller groups of images. Sometimes they trained the computer with only 50 images of healthy cells and 50 images of AML cells. This is important because in real hospitals, data can be limited.

They also simplified the images. Instead of using every tiny pixel, they measured basic features. For example, they looked at how bright the cell was, how rough its surface looked, and what shape it had. This is similar to describing a car by its color, size, and shape instead of every bolt and screw.

Two new ways of learning

The first method is called equilibrium propagation. You can imagine this like a marble rolling into a bowl. The marble naturally settles into the lowest point. The computer works the same way. It slowly settles into a stable state based on the input image.

Then the system gives itself a small push toward the correct answer. Over time, these small pushes help it learn. This method does not rely on backpropagation and works well with physics based systems.

The second method uses a small quantum circuit with just four qubits. A regular computer still runs the calculations, but the setup matches what real quantum computers can do today. A normal computer helps adjust the settings, like tuning a radio until the signal sounds clear.

What did the study find

The researchers compared these methods with a standard machine learning system. The classical system performed the best when it had enough data. With many images, it reached about 98 percent accuracy.

Equilibrium propagation reached about 86 percent accuracy. This is impressive because it learned without using backpropagation.

The quantum circuit reached about 83 percent accuracy. What stood out was how steady it was. Even when trained with very few images, it still performed well. The classical system needed many more images to reach its best results.

Why this matters in real life

In medicine, collecting data is hard. Doctors must carefully label images, and rare diseases may not have many examples.

A system that learns well from small datasets could help in rural hospitals or in research on rare illnesses.

This study shows that quantum based learning may become useful when data is limited, even if it is not yet better than current systems.

Looking to the future

Quantum machine learning is still young. Real quantum computers still make errors, and this study focused on a simple yes or no problem. But the results are promising.

They show that new learning ideas can work on real medical data. As quantum technology improves, these methods may one day help doctors diagnose diseases faster and more accurately.

The study is published in arXiv.

Knowledge becomes useful when it is easy to understand and simple to apply. Our goal at usefultoolsandproducts.comis to break down complex topics and highlight topics that support better decisions in everyday life.