Artificial Intelligence for Blood Cell Analysis

Gemmo Team
Gemmo Team

New Sound Analysis API

Test our API for sound analysis

Millions of patients go for blood tests every day, whether for a routine checkup, to monitor a pre-existing condition, or come up with a diagnosis. The analysis and counting of the different blood cells, including red and white ones, are typically performed manually. This means that experienced laboratory technicians examine each blood sample individually. However, as in other industries, Artificial Intelligence (AI) is also about to disrupt diagnostics in haematology. In particular, Computer Vision technologies help specialists get more accurate results faster by automating part of the process. Thus, making it possible to provide patients with meticulous results in a short time.

What is blood cell analysis?

Before delving into a specific use case, let us explain what blood cell analysis is. As the name suggests, this procedure involves examining a blood sample to learn more about its characteristics. The sample is usually taken from an arm, finger, or earlobe vein; in some circumstances, bone marrow blood cells may also be analysed. This test allows doctors to detect, count, and classify the different types of blood cells. Moreover, experts can detect abnormal cells if present, opening the door for early blood disease diagnosis.

Blood cell analysis is one of the most frequent tests prescribed by physicians to monitor patients’ health status. Abnormal values can be a flag for conditions such as platelet deficiency, infections, different forms of leukaemia, anaemia or coagulation disorders. On the contrary, if the values are in the normal range, the bone marrow can produce blood cells correctly, and no infections are present in the body.

How can AI be used to detect leukaemia in blood cells?

By the term leukaemia, we collectively refer to blood cell cancers. This group of cancers typically starts in the bone marrow and results in a high number of abnormal blood cells. Based on the type of blood cells affected and the cancer progression speed, we distinguish various types of leukaemia. In any case, early detection and diagnosis are key to a good chance of a cure. 

It is paramount to identify malignant cells with high accuracy to reach a correct diagnosis. However, distinguishing abnormal cells from normal ones under the microscope is a challenging task. In fact, the morphological images of the two cell types appear very similar. This is where Artificial Intelligence comes into play. Indeed, not only can Computer Vision systems help pathologists and oncologists make quicker and data-driven inferences, but they could also detect details that might otherwise escape the human eye.

What is the solution we applied in this case?

To assess the feasibility of employing an AI solution for blood cell analysis and leukaemia diagnosis, we studied the case of Acute Lymphocytic Leukaemia (ALL). This is a particular type of leukaemia characterised by the rapid development of a large number of lymphoblasts (i.e., immature white blood cells). The aim is to detect whether a white blood cell is abnormal or not so that cancer can be diagnosed. The solution we implemented is based on an artificial neural network. Specifically, it is a Convolutional Neural Network (CNN). This particular Deep Learning architecture – inspired by the biological processes within the animal visual cortex – can analyse the content of images. In practice, by processing the data contained in the pixels, the network learns its way of interpreting images and automatically extracts the distinctive features of the blood cell.

During our study, we first performed a segmentation of the microscopic images. In this way, we obtained multiple images, each containing a single blood cell. Then, we trained the network to distinguish normal from malignant cells in inference. 

What are the final results of this study?

Our study resulted in an accuracy of 99%. This means that, out of 100 cells, the system misclassified only one normal cell as malignant or vice versa. This shows how, despite the task’s difficulty due to the visual similarity between normal and abnormal cells, the latest advances in Artificial Intelligence allow for excellent results. Moreover, the results are reliable, and the blood analysis AI system is scalable and can operate in a matter of seconds. This means that clinics will perform tests faster and at a lower price while maintaining the same quality.

Nonetheless, it is important to emphasise that no Machine Learning system is perfect. In a critical field such as medicine, an expert must always assess all outcomes. Thus, the ultimate goal will never be to replace physicians and specialists with AI technologies. Rather, Artificial Intelligence must become an indispensable tool in the healthcare sector, assisting clinicians in achieving their main objective, which is to give patients the best care possible. 

Author Federica Baldi

Start your Ai journey today!