This app has taken most of my time in the last few months, which is why we didn’t publish any other articles.

We have created an application that can detect breast cancer on mammograms. Why is this interesting?

Because with the help of this application we can save human lives.

In cooperation with Dr. Srđan Jerković (and his company Medicus) and the General Hospital Pula, EXACT BYTE d.o.o. developed an application radiologists will use to help them find benign and malignant tumors on the breast.

Malignant tumors are what a layman would call cancer. The application itself is more accurate than the average radiologist. With the help of artificial intelligence  research developed by people around the world, we can see how much more accurate radiologists are using such an app.

If the average radiologist uses this program, will find additional 6 women per 1000 examinations that they would otherwise miss. On average, per 1000 examinations we will save one human life.


The rough amount of examinations in General Hospital Pula is 500 examinations per month, so, assuming that all radiologists are average (compared to the world) and that all radiologists use this program, an additional 6 women a year will survive and “cure” cancer.

Maybe 6 people in Pula, a city with 60,000 inhabitants, doesn’t seem a lot. Until someone in your family is among them. Until it’s your grandmother, mother, or sister.

These data are extrapolated from existing studies, and it should be taken into consideration that it is necessary to make a real study with such a program to obtain accurate data.

I can’t even imagine a better usage of artificial intelligence. Saving human lives and helping those in need is imperative in today’s world, regardless of politics, money, and power.

Politics, money, and power are things that should be secondary when we talk about human lives!

Deadly disease

In people under the age of 65, cancer is the most common cause of death, causing almost half of all deaths.

In Croatia, according to the latest available data, a total of 13,809 people died of invasive cancer in 2018, of which 8,049 were men and 5,760 women. In only one year!

The most common malignant causes of death in men were lung cancer (2,097 deaths), colon and terminal bowel cancer (1,321), and prostate cancer (772), and in women, colon and terminal bowel cancer (919), lung cancer (860), and breast cancer (789). The number of cancer deaths is on the rise, as is the proportion of cancer deaths (of all deaths).

Survival data shows that Croatia is (for the majority of malignant diseases) at the bottom of the European countries included in the study, with better results for the survival of malignant diseases in children.

There is a clear improvement in cancer survival, but similar to mortality, survival is improving faster in most other comparable European countries.

Reference – Croatian Institute of Public Health

Artificial intelligence is used in many fields, more and more, and has begun to be used in medicine as well. We hear about cars that drive on their own and fantasize about a day when we won’t have to drive anymore.

Now we have those same tools to help people in medicine, using artificial intelligence.

AI in practice

EXACT BYTE has toyed with the idea of ​​self-driving cars and we are waiting for the COVID situation to calm down a bit to present the Mini Tesla.

I would not enter such a car (it’s too small for me) for another 5-10 years because I know that there are problems with such programs and that these programs don’t have a “common sense”.

Programs in general, like computers, are pretty “stupid”.

I leave the metaphysical debates about whether Consciousness is just a “sufficient amount of calculation” for the pub after a few beers, but the fact is that people can “compute” a lot and we take it for granted, as well as our “common sense”.

We mentioned something more about that as well, in an article on recognizing people using artificial intelligence.

In a specific example of this program that recognizes breast cancer, it can tell what the probability is that there is a malignant or benign growth on the mammogram image, and it can mark it on the image.

If you insert pictures of cats or dogs instead of mammograms for analysis, it will continue to look for breast cancer, although it is obvious to every human that the picture of dogs and breasts is not the same thing.

I don’t want to go into details about how this can be resolved, but it is important to say that it is necessary to have a separate program that only recognizes breast images to “understand” it.

In other words, such programs are very limited, and only do one thing well. It is necessary to have a doctor who interprets these pictures and who makes the conclusions!

Artificial intelligence is a topic that has been covered in detail and there is this grandiose image is being constructed, while its foundations are relatively simple. These are programs that “learn by themselves” based on examples.

Programs that learn on their own

How does this “learning” work? In this case, we give the algorithm, which can be considered a small program, images of mammography and tell it which image is “cancer”.

The mammogram image consists of 4 “views” which means through one mammograph we get 4 separate pictures. The standard images obtained from mammography are “bilateral craniocaudal” (CC) and “mediolateral oblique” (MLO).

We take one image for CC on the left – L-CC and one on the right – R-CC.

Also, we take one image for MLO on the left – L-MLO and one on the right – R-MLO.

There is one relatively simple way to see the difference between CC and MLO – in MLO images we often see the part of the arm (armpit) at the edge of the image.





After a lot of examples (we are talking about thousands of images), the program corrects itself and learns to recognize benign and malignant tumors.

Once such a program gets those same 4 images, it can produce the probability that it has found a benign and/or malignant mass and can show where it found that same mass on each of the “views”. Example of R-CC in the figure.


A visible malignant tumor that might look suspicious to the untrained eye – green and orange are the results of the program, the image is originally black/white.

Programs that save lives

The program is based on the existing architecture that has been proven in the research. Some small differences in the network and details about its implementation will be revealed when the program itself stabilizes and proves to be reliable, but I want people to understand that the foundation of this program is the result of research on people in the field of artificial intelligence around the world. It’s not an idea from a single developer based in Pula.

Thanks to the cooperation with Dr. Jerković and General Hospital Pula, we were able to access anonymized images of the mammography patients. These are images of patients about whom we know nothing, we only see images of the patient’s mammography, to maintain and respect their privacy.

Based on these images, as previously explained, we give the program 4 mammography images and send them for analysis. The results of the analysis help doctors make better decisions and perhaps consider certain parts of the images that they did not immediately see as significant.

We have to keep in mind that people get tired and after hours and hours of viewing images, fatigue can play a role, while one program does not get tired and works just as well after 10 days as the first hour of viewing images.


Programs that save human lives and help people live better and better should be the programs we develop every day. These should be programs in which the largest sums of money are invested, and you should constantly look for ideas for other programs that would help people.

The program we created is very precise in the analysis of mammography images, but it requires a doctor who will take into account the offered analysis and make a final decision based on many more factors than the image itself.

This program was developed and put to use completely free of charge.

The program is the result of cooperation with Dr. Srđan Jerković and OB Pula, without which this would not be possible. I’m taking this opportunity to thank them for their cooperation.

The doctors I met, as well as doctors in general, are people who work through the long days and nights to give people a chance to overcome illness and live a better life. These same doctors are people of flesh and blood, often on the verge of fatigue.

We take them for granted because we are used to always having them here.

People don’t think things were very different 100 years ago, let alone a few hundred years ago and how things would be if we didn’t have those same doctors.

People talk about how something needs to change. How bad our country is and how our health care systems do not work. But rarely does anyone take action and try to fix it. Let’s leave the brilliant stories about what we need to change behind. If we don’t take matters into our own hands and try to make this world a better place, no one will do it for us.

I sincerely hope that more hospitals and interested institutions across the world will join in using this free program to improve it and have the opportunity to help as many people as possible.

In the future, we will publish a more professional article that will contain more technical data and in which people who are interested in details will be able to find their answers.

If you are interested in any of these details, contact us so that we can answer not only you but also other readers who are interested in your question.