Los Angeles, USA – When echocardiography reveals left ventricular hypertrophy, the doctor faces a tedious and time consuming task: to determine its extent and cause. Artificial intelligence (AI) can make this easier, as a US study has shown that not only can the geometric features of the ventricle be accurately measured but also the underlying conditions including aortic stenosis, hypertrophic cardiomyopathy and cardiac amyloidosis. The work has been published Clothing network.
The team highlights this research Grant Duffy (Cedars-Sinai Medical Center in Los Angeles): A large set of data from medical videos has never been published before. “With nearly 24,000 cardiac ultrasound videos, this is an impressive publication, if only for the huge amount of data used by the authors – such as training, verification of their algorithms and testing – at different stages of their work”, confirms Dr. Jackie’s motherInterview with Leader of Applied Machine Learning Group at Fraunhofer Heinrich-Hertz Institute, Berlin Medscape
AI saves doctors a lot of time and effort
If Jackie Ma sees the importance of AI Pr David Wang, The last author of the study put it back in its place: “It is important to note that algorithms leave the decision to physicians. They simply offer to help with this or that diagnosis, ”reads the TCTMD Cardiology website.
Internationally, Germany has a fairly strong position in AI research, development and applications, reports Jackie Ma. Grant Duffy, Rather to analyze the ECG. He also developed algorithms for proteins and genomes and EEG and cerebral MRI, chest X-rays and specific tumor diagnoses. The researchers were also interested in the Kovid-19 epidemic by researching the personal risk of contamination, depending on the distance and time spent with an infected person.
An important point: the acquisition of reference data
Jackie Ma says, “We work closely with different clinics, but strict data protection makes it difficult to access the appropriate source material. The authors of the study share the same experience: ” One of the major challenges in using AI in healthcare is the lack of benchmarks. A
However, the authors are fortunate to have been able to source data from their own clinic as well as the Stanford Center, one of the specialties of which is Cardiac Hypertrophy. They captured videos of the long parasternal axis on one side, on which one could see the two ventricles, the main and left atrium of the aorta, and on the other, images showing 4 cavities from the top.
Echocardiography, for first choice diagnosis
According to the recommendations of the educated society, echocardiography is the most widely used method of diagnosing hypertrophy, the study authors explain. Jackie underscores the problems with motherhood: “Because of the sheer amount of information they provide, it is difficult for AI users to manage echocardiograms: they require a huge investment in time, power computing and storage space.”
Grant Duffy and his colleagues used a portion of the videos to train their algorithm, after each passage, indicating whether it had reached the correct or incorrect result (see box). The analysis report and the attached comments each time served as a reference. “The question is always: where does the analysis and the vaccine come from? It would be ideal for doctors to give their opinion as much as possible,” said Jackie Ma.
This procedure will be especially useful for patients with hypertrophy because even experts find it difficult to distinguish between different pathologies that alter the heart in a morphologically similar way. What is confusing is that the symptoms vary from person to person and are sometimes mild, sometimes more severe. In addition, the measured values vary due to filling time and irregular heartbeat.
Yet a reliable test is important, as it determines the continuity of the procedure due to its prognostic importance. This way it is possible to assess the risk of sudden cardiac death and determine which patients need a defibrillator. In addition, classification based on genetic differences will also be possible.
An algorithm that increases the diagnostic potential of ultrasound
“So we tried to find out if the hidden potential of echocardiography could be brought out by combining it with artificial intelligence, deep learning (see box),” Grant Duffy and his team explained to justify their previous interests. Study.
In a possible comparison with two trained cardiologists, the algorithm showed somewhat better results.
The method was successful: the algorithm identified the deviations, even when they were subtle. For intraventricular wall thickness, the mean error was only 1.2 mm. It was 2.4 mm for left ventricular diameter and 1.4 mm for posterior wall thickness. In a possible comparison with two trained cardiologists, the algorithm showed somewhat better results. It performed equally well when capturing data from other US clinics and other countries. According to researchers, this proves that its relevance continues across continents and across various health systems.
A process that mimics the visual cortex
Investigators traced the primary condition using a special deep learning method called visual cortex, Convulsive neural network. This “convoluted neural network” was thus able to differentiate between high-level confirmed cardiac amyloidosis, hypertrophic cardiomyopathy, and aortic stenosis, among other possible causes.
The similarity between these underlying conditions is that they create a chronic overload that the heart tries to overcome by rebuilding.
In many patients, the disorder is of systemic origin: the heart muscle has to work against increased pressure, for example, in the case of aortic stenosis, but especially in the case of hypertension, which begins to rebuild in 60% of patients.
Trigger genetic defects in muscle fibrils
But thickening can also be caused by pathological processes within the heart, such as in hypertrophic cardiomyopathy. The latter finds its source in hereditary mutations: in more than 1,500 gene coding known primarily for sarcomere proteins.
Diagnostics must differentiate between another series of diseases that directly weaken the heart tissue: amyloidosis. Only a few of them develop due to inherited genetic defects. Most are caused by disease of the bone marrow or lymph nodes or by chronic inflammation such as rheumatoid arthritis, Crohn’s disease or ulcerative colitis. As a result, misfolded protein fibrils accumulate in the cell. To date, about thirty proteins have been discovered that cause amyloidosis.
Finally, if their presence is rare, other common congenital disorders, such as Fabry’s disease, Friedrich’s ataxia, or Melas’s syndrome, are not ruled out.
A rare disease, but probably more common than you think
Doctors can easily miss the initial condition because it presents it as a consequence of “normal” hypertension or “normal” kidney disease, explains David Oiang in tctMD. “Cardiac amyloidosis probably deserves the paradoxical title of ‘Common Rare Disease’ because it is broader than the published figures.”
Hypertrophic cardiomyopathy also presents a diagnostic challenge due to its lack of uniformity: it sometimes spreads, with walls 60 mm thick, but sometimes only the peripheral areas are slightly thicker.
The authors consider their model to be a platform technology because it is suitable for screening for wounds caused by all kinds of diseases such as valvulopathy or chemotherapy. David Oiang added, “Finding patients who have not been diagnosed in routine practice can be done subtly.”
The Acquiring deep knowledge
A sub-domain of deep learning Machine learning. Although the latter’s algorithms run through the tree of mathematical decision, the dive of deep learning, so to speak, goes deeper. It is based on the neural network whose operations are inspired by the human brain. These networks consist of an arrangement of input and output neurons, connected by a variable number of intermediate layers.
Thus the algorithms constantly combine new content and take other paths to develop the ability to independently determine the criteria for accurate identification. Unlike machine learning, programmers no longer interfere with these processes: they simply feed them with basic information, from which algorithms make predictions and make decisions. In practice, these are suitable for looking for patterns, for example image analysis but also outside of medicine, for recognition of face, object or speech.
The calculations lead to fully automated and precise and reproducible measurements.
This article was originally published on Medscape.de and had the title “Beindruckend” – Consultant Intelligence Machines Cardiologic Concurrence . Translation / Adaptation by Dr. Claude Leroy.
Follow Medscape in French Twitter.
Follow theheart.org | Medscape Cardiology introduced Twitter.