Shutterstock / Prostock-studio ” src=”https://s.yimg.com/ny/api/res/1.2/I16qPXYdtjNfduJIlP0_ZA–/YXBwaWQ9aGlnaGxhbmRlcjt3PTk2MDtoPTYzOS4zMzMzMzMzMzMzMzM0/https://s.yimg.com/uu/api/res/1.2/5pW79qhnURR70z5IZD0svQ–~B/aD05NTk7dz0xNDQwO2FwcGlkPXl0YWNoeW9u/https://media.zenfs.com/es/the_conversation_espa_a/9e5d99bd5cb1d305c2683efc48e757a3″ data-src=”https://s.yimg.com/ny/api/res/1.2/I16qPXYdtjNfduJIlP0_ZA–/YXBwaWQ9aGlnaGxhbmRlcjt3PTk2MDtoPTYzOS4zMzMzMzMzMzMzMzM0/https://s.yimg.com/uu/api/res/1.2/5pW79qhnURR70z5IZD0svQ–~B/aD05NTk7dz0xNDQwO2FwcGlkPXl0YWNoeW9u/https://media.zenfs.com/es/the_conversation_espa_a/9e5d99bd5cb1d305c2683efc48e757a3″/>
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease of irregular and asymmetric progression, characterized by a progressive loss of motor neurons that leads to muscle atrophy, paralysis and finally death. The life expectancy of these patients is 3 to 5 years from the onset of symptoms.
There is currently no cure for ALS, but early detection can slow progress. In this sense, it must be distinguished that not all patients with ALS are the same. The disease is known as spinal ALS (80% of cases) when the first symptoms appear in the arms and legs (beginning in the extremities or in the spine). And we speak of bulbar ALS (20% of cases) when it begins in the medulla oblongata (bulbar initiation).
Patients with the later form tend to be shorter-lived due to the critical nature of the function of the bulbar muscle responsible for speech and swallowing. However, 80% of all ALS patients experience unclear and difficult speech articulation (dysarthria, in medical jargon). On average, speech does not begin to show signs of worsening until about 18 months after the first bulbar symptom appears.
These symptoms are usually noticed early in bulbar ALS disease or in the later stages of spinal ALS. Early identification of bulbar disease in people with ALS would be critical to improving diagnosis and prognosis and may be the key to effectively slowing down the disease.
The key is in the vowels
The bad news is that, at the moment, there are no standardized diagnostic procedures to evaluate bulbar dysfunction in ALS. The good news is that it is possible to detect early, often imperceptible, changes in speech and voice through objective measurements as suggested in previous work.
From the Distributed Computing group of the University of Lleida and the International Center for Numerical Methods (Barcelona) we have shown that bulbar disease can be detected early using acoustic parameters obtained by analyzing vowels.
At this point, it is interesting to know that in artificial intelligence (AI) it is necessary to obtain the characteristics, properties or differential signs to be able to perform classifications. This is achieved using machine learning algorithms, which is the field of AI used in this ALS research.
Machine learning tries to classify, guess and predict diseases, weather, earthquakes, stock market fluctuation, price evolution, demand, etc. In our case, we apply it to the diagnosis of a disease, the bulbar disease in ALS patients, taking into account the voice characteristics of an individual. The question we asked ourselves to carry out the research was basically two: what characteristics to choose and what sounds. Answering them was the main challenge of this research.
Sounds become signals. These can be processed by a computer to obtain the characteristics. In this experiment it was demonstrated that the most important characteristics to identify the bulbar affection were the fluctuation, brightness, harmonic-noise relation and tone of the vowel sounds.
Machines can perceive more sounds than specialists
The test was carried out with 45 patients with ALS and 18 participants without it, necessary to be able to make comparisons. Regarding the sounds, the vowels (in Spanish) were chosen, since they are the most important in the speech of any language.
Once the characteristics of the patients who participated in the study were obtained, we used several machine learning algorithms. The SVM algorithm offered the highest performance, obtaining an accuracy of 95.8%. That is, it detected with 95.8% if a participant in the study had a bulbar disease.
Another interesting result was the fact that, in some cases, the machine learning models outperformed the specialist’s diagnosis. After all, humans are not capable of perceiving sounds that machines can.
The results obtained are very encouraging and show that we may be in front of an adequate tool to help multidisciplinary clinical teams to improve the diagnosis of ALS.
This article was originally published on The Conversation. Read the original.
Francesc Solsona Tehas receives funds from the Ministry of Economy, Industry and Competitiveness (TIN2017-84553-C2-2-R).