Technique can be used in the characterization, diagnosis and treatment of neurological diseases
The development of an automated method to analyze lesions in the white matter of the brain can assist doctors and specialists in the characterization, diagnosis and treatment of neurological and psychiatric diseases. The technique was developed and described by researcher Mariana Pinheiro Bento as part of her doctoral thesis defended at the Faculty of Electrical and Computer Engineering (FEEC) at Unicamp.
The study evaluated more than 350 patient images. There was collaboration with Canadian researchers from the University of Calgary. There is no interface available yet for using the methodology. The research that resulted in the method was guided and co-supervised, respectively, by professors Letícia Rittner and Roberto de Alencar Lotufo, both from FEEC.
Lesions in the white matter of the brain are observed in patients with different diagnoses, especially autoimmune diseases, such as multiple sclerosis, systemic sclerosis and lupus. Mariana Bento explains that the analysis of brain injuries currently used is done manually using magnetic resonance images. “This represents a non-trivial, costly and subjective task.”
The author of the study informs that there are studies reporting automated analyses, but focused on just a single disease. “Diseases that generate lesions in the white matter have distinct characteristics. In the work we are aware of, automated analysis is focused on just one type of disease and typically uses a small set of images. Our goal was to be more generalist and robust. Therefore, our method works with injuries from patients with different diagnoses”, she compares.
Machine learning
The method described makes use of machine learning (machine learning algorithm ), field of computer science that explores the study and construction of algorithms that can “learn” and make predictions about data, in this case, images.
“We didn’t program the machine to analyze an injury. We program the computer to 'learn' how to analyze an injury. This is done based on various information and characteristics about the injuries that doctors and specialists gave us.”
Still according to Mariana Bento, image processing and pattern recognition techniques were combined. “Based on the magnetic resonance images, the computer tries, through a learning process, to extract important characteristics from these images and point out the abnormal and healthy tissues of the brain.”
How is done
There are three stages in the automatic lesion analysis process: detection, segmentation and comparison. First, the method informs whether or not the patient has a brain abnormality, simply by answering that question. The second stage consists of segmenting, that is, identifying the location of the anomaly and the size of the lesion. There is also a third stage. This is a longitudinal analysis that evaluates, over time, how the lesion is changing: whether it changes shape, size, texture, etc.
The research that resulted in the description of the methodology included the collaboration of researchers from the area of rheumatology at the Faculty of Medical Sciences (FCM) and the Brazilian Institute of Neuroscience and Neurotechnology (BRAINN). There was financial support from the São Paulo State Research Support Foundation (Fapesp) and the Coordination for the Improvement of Higher Education Personnel (Capes).
