{"id":1906,"date":"2025-06-02T15:28:40","date_gmt":"2025-06-02T13:28:40","guid":{"rendered":"https:\/\/kleefstrasyndrome.com\/?p=1906"},"modified":"2026-08-20T16:43:21","modified_gmt":"2026-08-20T14:43:21","slug":"machine-learning-algorithms-in-eeg-of-kleefstra-syndrome","status":"publish","type":"post","link":"https:\/\/kleefstrasyndrome.com\/en\/machine-learning-algorithms-in-eeg-of-kleefstra-syndrome\/","title":{"rendered":"EEG and Artificial Intelligence in Kleefstra syndrome: what we know and what the next steps are"},"content":{"rendered":"

In 2025, it was published in the scientific journal Sensors<\/em> my review entitled \u201cMachine Learning Algorithms in EEG Analysis of Kleefstra Syndrome: Current Evidence and Future Directions\u201d<\/em>. The paper brings together what we know to date about the electroencephalogram (EEG) in Kleefstra syndrome and examines how modern signal analysis and machine learning methods could be utilized in the future to better study brain function in the syndrome.<\/p>\n\n\n\n

What do we know today about the EEG in Kleefstra syndrome?;<\/h2>\n\n\n\n

The electroencephalogram is a non-invasive test that records the electrical activity of the brain. In Kleefstra syndrome, it is mainly used when there are epileptic seizures or other neurological symptoms. Clinical reports published to date show that approximately 20\u201330% of individuals with Kleefstra syndrome experience seizures<\/strong>. Different types of seizures have been described, such as generalized tonic-clonic seizures, absence seizures, and focal seizures. However, electroencephalographic findings show significant heterogeneity. To date No consistent EEG pattern or electrophysiological biomarker specifically characterizing Kleefstra syndrome has been identified<\/strong>. This means that the EEG is not a diagnostic test for the syndrome and cannot replace genetic testing.<\/p>\n\n\n\n

Limited knowledge is not only due to the rarity of the syndrome. The available EEGs mainly come from isolated clinical cases or small series of patients, often with different recording protocols, different ages, and different clinical pictures. This makes it difficult to compare results between studies.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Beyond the simple visual assessment of the EEG<\/h2>\n\n\n\n

In daily clinical practice, the EEG is evaluated primarily visually by specialists. However, modern research can extract much more quantitative information from the same signal.<\/p>\n\n\n\n

\u0397 quantitative EEG (qEEG)<\/strong> it can, for example, study the power of different brain signal frequencies, the functional connectivity between different brain regions, as well as more complex signal properties, such as the aperiodic background of EEG activity. <\/p>\n\n\n\n

Such approaches have already been used in other neurodevelopmental disorders. Syndromes such as Angelman, Rett, and Fragile X<\/strong> They have been studied more extensively electrophysiologically and offer useful examples of how quantitative EEG biomarkers can be investigated in rare genetic disorders. <\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Where can Artificial Intelligence contribute?;<\/h2>\n\n\n\n

Machine learning algorithms can process a large number of features simultaneously and search for combinations or patterns that are difficult to identify through visual observation of the EEG alone.<\/p>\n\n\n\n

In a future study on Kleefstra, features such as brain rhythm power, the frequency of epileptic discharges, or measures of brain connectivity could be analyzed, for example. Then, classical algorithms like Support Vector Machines<\/strong> or the Random Forests<\/strong> they could investigate whether any combination of these features separates different clinical groups. Such methods are of particular interest in rare diseases because they can also be used in relatively small samples, provided the analysis and validation are carefully designed. <\/p>\n\n\n\n

Deep learning represents yet another potential direction, but it requires much larger volumes of data. In such a rare syndrome, directly training large neural networks on a small number of EEGs carries a significant risk of overfitting. Therefore, approaches are being considered such as transfer learning<\/strong>, where a model is initially trained on a large EEG database and then adapted to a smaller dataset, as well as techniques data augmentation<\/strong> to address the data shortage. <\/p>\n\n\n\n

For now, however, these possibilities in Kleefstra syndrome remain research<\/strong>. The most significant difficulty is not the lack of available algorithms, but the absence of large, standardized, and publicly available EEG datasets specifically for the syndrome. <\/p>\n\n\n\n

From an idea to an international EEG database<\/h2>\n\n\n\n

One of the key conclusions of the review is that before we can seriously talk about Artificial Intelligence applications in Kleefstra, the appropriate research infrastructure must first be established.<\/p>\n\n\n\n

The paper proposes as main directions:<\/p>\n\n\n\n