Among geriatric patients, dementia and depression are the most common disorders seen in psychiatric consultations at general hospitals. According to the latest estimates, the number of people with dementia in Germany will increase significantly by 2050. While just under 1.6 million people in Germany were living with dementia in 2018 (1.9% of the population), experts at Alzheimer Europe predict that the number will rise to 2.7 million (3.4%) by 2050. The annual cost of dementia is estimated at 5.6 billion euros, more than half of which is accounted for by care costs. In Germany, 8.7% of people over 65 also suffer from depression, which often accompanies dementia.

Depression and dementia influence each other. Depression increases the risk of developing dementia by up to six times. This risk is higher than for other chronic diseases. Conversely, the risk of developing depressive disorders is also significantly higher in people with dementia. Depressive disorders also impair cognition, activities of daily living (ADL), and social skills in dementia patients, making their symptoms appear to be even more pronounced. Both clinical pictures pose great challenges for patients and their relatives.

Due to their usually gradual onset, dementia is in many cases identified too late. Likewise, diagnosis and treatment of depression in older people often take place late or not at all, as people assume that the symptoms they show are normal signs of aging. The symptoms that occur complicate diagnosis because they differ from the depressive symptoms shown by people in younger age groups: physical impairment, memory problems, sleep disturbances, inner restlessness, and irritability are mentioned more frequently.

As memory problems are usually the first symptoms of dementia, the first memory consultations were established a little over 20 years ago to test memory and other mental functions in a professional capacity. Early diagnostic clarification with the resulting possibility of rapid initiation of treatment should delay the progression of dementia and thereby lead to an improvement in quality of life for the patient and their caregiving relatives. Around a third of patients referred to specialized centers for a dementia diagnosis suffer from a depressive disorder, and 30-50% of patients with Alzheimer’s exhibit depressive symptoms, especially in the early and middle stages. For this reason, the memory consultation also includes a check for depressive symptoms in patients.

Due to the correlation between dementia and depression, together with their overlapping symptoms, the diagnosis of whether depression or dementia is the primary disease or even whether both are present is not always clear. To assist therapists here, our research explores methods based on natural language processing (NLP), automatic speech recognition (ASR), and deep learning (DL) to distinguish between dementia and depression and provide an additional indication for diagnosis. Since a diagnosis during the memory consultation is based on the implementation and evaluation of a standardized screening procedure, we record the conversations between patients and therapists during the screening sessions. This provides us with a dataset that includes free speech from a semi-standardized interview and provoked speech from two standardized test batteries (SKT and CERAD). On the basis of this dataset, we are working on test automation using NLP and ASR to predict test scores, for example.  We are also investigating different approaches that extract linguistic and acoustic features from audio to learn patterns for dementia and depression. Examples of linguistic features include word repetitions, pauses in speech, and sentence breaks; examples of acoustic features include intonation, speech speed, and speech volume. Based on these and many other features, we train neural network architectures with DL algorithms. Classification and regression methods are prediction methods from supervised learning, which are then used to discriminate between dementia and depression based on speech.

Some approaches already exist to investigate the identification of depression or dementia using language and DL approaches [1][2][3][4][5][6][7][8][9][10]. However, these works exclusively compare patients suffering from either dementia or depression against healthy control subjects. We are aware of only a few papers contrasting the two diseases [11][12]. In our eyes, as the problem of discriminating between the two has been a well-known challenge in psychotherapy for a long time, it cannot be neglected.

References

• [1] Pérez-Toro, P. A., et al. “Influence of the Interviewer on the Automatic Assessment of Alzheimer’s Disease in the Context of the ADReSSo Challenge.” Interspeech 2021, ISCA, 2021, pp. 3785-89, https://doi.org/10.21437/Interspeech.2021-1589.
• [2] Toth, Laszlo, et al. “A Speech Recognition-Based Solution for the Automatic Detection of Mild Cognitive Impairment from Spontaneous Speech.” Current Alzheimer Research, vol. 15, no. 2, Jan. 2018, pp. 130-38, https://doi.org/10.2174/1567205014666171121114930.
• [3] Themistocleous, Charalambos, et al. “Identification of Mild Cognitive Impairment From Speech in Swedish Using Deep Sequential Neural Networks.” Frontiers in Neurology, vol. 9, Nov. 2018, p. 975, https://doi.org/10.3389/fneur.2018.00975.
• [4] Luz, Saturnino, et al. “Detecting Cognitive Decline Using Speech Only: The ADReSSo Challenge.” ArXiv:2104.09356 [Cs, Eess], Mar. 2021, http://arxiv.org/abs/2104.09356.
• [5] Luz, Saturnino, et al. “A Method for Analysis of Patient Speech in Dialogue for Dementia Detection.” ArXiv:1811.09919 [Cs, Eess], Nov. 2018, http://arxiv.org/abs/1811.09919.
• [6] Fraser, Kathleen C et al. “Linguistic Features Identify Alzheimer’s Disease in Narrative Speech.” Journal of Alzheimer’s disease: JAD, vol. 49,2, 2016, pp. 407-22, https://doi.org/10.3233/JAD-150520.
• [7] Villatoro-Tello, Esaú, et al. “Late Fusion of the Available Lexicon and Raw Waveform-Based Acoustic Modeling for Depression and Dementia Recognition.” Interspeech 2021, ISCA, 2021, pp. 1927-31, https://doi.org/10.21437/Interspeech.2021-1288.
• [8] Liu, Zhenyu, et al. “Ensemble-Based Depression Detection in Speech.” 2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), IEEE, 2017, pp. 975-80, https://doi.org/10.1109/BIBM.2017.8217789.
• [9] Chlasta, Karol, et al. “Automated Speech-Based Screening of Depression Using Deep Convolutional Neural Networks.” Procedia Computer Science, vol. 164, 2019, pp. 618-28, https://doi.org/10.1016/j.procs.2019.12.228.
• [10] He, Lang, and Cui Cao. “Automated Depression Analysis Using Convolutional Neural Networks from Speech.” Journal of Biomedical Informatics, vol. 83, 2018, pp. 103-11, https://doi.org/10.1016/j.jbi.2018.05.007.
• [11] Sumali, Brian, et al. “Speech Quality Feature Analysis for Classification of Depression and Dementia Patients.” Sensors, vol. 20, no. 12, June 2020, p. 3599, https://doi.org/10.3390/s20123599.
• [12] Mohammad Abdallah_Qasaimeh, Bashar, et al. “Detecting Depression in Alzheimer’s and MCI Using Artificial Neural Networks (ANN).” International Conference on Data Science, E-Learning and Information Systems 2021, Association for Computing Machinery, 2021, pp. 250-253, https://doi.org/10.1145/3460620.3460765.