Medicine and Healthcare
Measuring the Heart
Artificial intelligence Determines Prognosis for ToF Patients
Scientific name of the study
Prediction of prognosis in patients with tetralogy of Fallot based on deep learning imaging analysis
The number of adults with congenital heart disease is growing constantly. Most patients’ survival is ensured by modern surgical techniques today. This also applies to patients with tetralogy of Fallot. Tetralogy of Fallot (ToF) is one of the most frequent congenital cyanotic heart defects. It affects about 12.5 percent of all adults with congenital heart disease.
After corrective surgery, these patients’ long-term prognosis is assumed to be good. This is especially the case in the Western industrial countries. However, late complications such as pulmonary valve insufficiency or heart rhythm disorders originating from the heart chambers all the way up to sudden cardiac death must be anticipated. As a result, those affected require lifelong specialized medical follow-up after the corrective surgery.
The further development of the heart and pulmonary function strongly depends on the further course of the corrected organ. In this context, what matters most are the heart chambers, that is, their structure, the structure of their tissue and the resulting pump function. Today, modern imaging techniques such as cardiac magnetic resonance imaging (MRI) facilitate a largely reliable assessment of the parameters in question. These examinations have to take place regularly to make sure that appropriate medical treatment is initiated in time, where needed. Those affected rely on short waiting periods and care provided from one source. This is an increasing challenge for those involved in the medical care of this growing patient group.
How does Artificial Intelligence Benefit the Prognosis Assessment?
Is it possible to use artificial intelligence (AI) for evaluating MRI images in such a way as to reliably predict potential ventricular tachycardia, as well as the risk of sudden cardiac death? And might this serve as a basis for enabling precise intervention? ACHD specialist Gerhard-Paul Diller and his research team at the University Hospital of Münster have now investigated this question. In a feasibility study, they used parameters that had been calculated by algorithms on the basis of a nationwide collection of medical data and archived MRI images of patients with corrected tetralogy of Fallot (ToF) for the first time.
In a Nutshell
Artificial Intelligence in Medicine
What Is it About?
Researchers hope that a targeted use of artificial intelligence will lead to a better accuracy of diagnoses, as well as reliable information on the long-term course of diseases. The aim is to improve treatment and treatment strategies, to heal as-yet incurable diseases or to at least improve the patients’ life expectancy and quality of life. The procedure in question here is the targeted application of machine learning. By specially programmed calculation processes, huge numbers of digital data, as well as data structures, which are medically significant, are analyzed. Based on that, diagnoses, long-term courses and treatment options are generated. In this context, artificial intelligence means the programmed capacity of the applied algorithms (computational processes) to detect, differentiate, classify and assess complex correlations.collapse
“We used the MRI recordings of 42 patients with corrected ToF at the University Hospital of Münster for training a multilayer network that was based on Deep Learning algorithms. Subsequently, we applied this to an external imaging dataset that had been collected prospectively by 14 German heart centers. Said dataset comprises the medical data and image data of 372 patients with corrected ToF,” Gerhard-Paul Diller says, explaining the procedure.
Good to Know
Deep Learning Algorithms
What Are They Capable of?
Deep learning algorithms play the major part in machine learning, which is one of the techniques most commonly used in artificial intelligence. DL algorithms are self-adapting computational processes, which means that they can adjust themselves without outside influence. They are capable of identifying and evaluating multilayer data structures, from which they then learn in order to solve specific tasks or problems.collapse
Long-term Observation Yielded Important Clues
In order to assess and compare methods, the scientists additionally used conventional methods of prognosis assessment. This test revealed that the AI-based prognosis model, which is significantly faster, was in no case inferior to the established conventional methods. “In the case of an adequate image quality, the AI-based measurements of the ventricular function and ventricle size yielded a reliable prognosis. This evidence suggests that this method should be developed further. AI-based prognosis assessments can be a crucial factor in finding a more efficient treatment for the patients,” Gerhard-Paul Diller states, summarizing the study results.
In this context, the results of a long-term observation yielded significant evidence. The study included 372 participants of the National Register and covered a period of ten years. Late complications were observed in 23 of the study participants. Sixteen patients suffered ventricular tachycardia. Seven patients suffered sudden cardiac death which they either survived thanks to appropriate resuscitation measures or due to which they deceased. On this basis, the researchers were able to identify appropriate parameters for the prognosis assessment.
Increased Risk Requires Specialized Treatment
As could be seen, the risk of late complications already mentioned increases in patients with an enlarged right atrium and a limited capacity of the right ventricle, as well as in patients with an impaired function of the right and left ventricle that is caused by an altered structure of the organ. In this light, again, the researchers urgently recommend providing medical care and treatment of those affected in centers that are specialized in treating adults with congenital heart disease.
Scientific Details of the Study
Learn more about the study design, material and methods, as well as the background of the study:
Prediction of prognosis in patients with tetralogy of Fallot based on deep learning imaging analysis.
Diller GP, Orwat S, Vahle J, Bauer UMM, Urban A, Sarikouch S, Berger F, Beerbaum P, Baumgartner H, German Competence Network for Congenital Heart Defects Investigators ,
Heart (British Cardiac Society) 106, 13, 1007-1014, (2020). Show this publication on PubMed.