In recent years there has been an increase in the use of applications, tools and solutions that use machine learning in the field of medicine. The solutions cover the prediction of diseases, imaging analysis to filter or select relevant cases and even provide extra information to doctors for better decision making.
In this text I will share a data science solution that includes a component of machine learning implemented in Spain. The purpose of this mechanism is to filter relevant cases so that cardiologists make better use of their time and experience.
An echocardiogram is similar to an ultrasound: it is a non-intrusive study that measures the strength, size, and shape of the heart to identify or rule out various conditions in a given patient. To quantify the ejection force of the heart, it is necessary to identify its four chambers, calculate their volume, segment them and/or delimit them.
The cardiologist is the only person responsible for performing and analyzing an echocardiogram, through a two-phase process: taking the echocardiogram in the presence of the patient and analyzing it. The first phase takes approximately 30 minutes or more, depending on the age of the patient; the second takes 20 minutes per study. However, the delivery of the results can take a couple of days due to the amount of work of the cardiologists.
Additionally, it is common for a cardiologist to spend about three hours a day sitting at a computer delineating objects in an image for segmentation. A segmentation error implies an error in the volume calculation, in the ejection force and probably in defining whether or not there is a problem in the heart.
The solution of machine learning The proposal focuses on being able to identify and delimit the four chambers of the heart, as well as calculate the volume and ejection force to generate a recommendation on whether it is necessary for a cardiologist to review the case, either because the model has doubts about whether there is a problem –due to the value of the ejection force produced– or because the model is sure that the study should be reviewed by a cardiologist.
This model facilitates the mechanized process of cavity identification and segmentation, in order to later be able to generate the volume calculation and the ejection force. So this simple solution allows cardiologists to free up hours of work so that they can care for more patients, make better use of their experience and invest this time in doing what they are passionate about. This example shows that the use of artificial intelligence with proper planning can have a positive impact on certain fields, such as medicine.