Blog posts
Machine Learning for dementia early detection
Published on 10 June 2022Dementia affects in its big majority elderly people and exists in 4 types demonstrated in the following proportions [1] :
- Alzheimer dementia (~60%) : Alzheimer diagnosed patients evolving towards dementia.
- Vascular dementia (~20%): mainly related to blood clots obstructing the blood vessels towards the brain, with risk factors such as arterial hypertension, diabetes, atherosclerosis, auricular fibrillation, high fat ratio, smoking, stroke episodes.
- Lewy bodies dementia (~15%): 40% of Parkinson disease estimated to evolve towards the following dementia type.
- Fronto-temporal (~5%): tau protein produced in excess causing atrophy of the frontal and temporal lobes, quite related to hereditary factor.
Dementia is usually revealed by a degradation of cognitive, neuropsychiatric and daily life activities functions. At the clinical level, it is therefore diagnosed via combinations of cognitive and neuropsychiatric assessments, such as the Clinical Dementia Rate (CDR) [2] for example, that provides a score used to classify and identify individual’s dementia profile; i.e. either normal, Mild Cognitive Impairment (MCI) or established dementia. Other assessments exists that also integrate the evaluation of daily life activities, such as the Alzheimer’s Disease Cooperative Study-Activities of Daily Living (ADCS-ADL) [3]. Some clinical procedures even take it a step further and use the following metrics for driving identified patients towards a final MRI for definitive diagnosis.
The aforementioned clinical procedures however require in most cases direct interaction with the caregiver. They might therefore be very time consuming and represent a heavy burden for specialists, refraining evidently from early detections among majority of people. Prompt dementia identification had therefore become a major challenge in the last decades, provided the offset between symptoms appearance and sickness evolution. The advent of Machine Learning (ML) techniques in that sense represent a crucial advantage for the scientific community and majority of recent research projects are evolving towards this direction for dementia early identification.
As a matter of fact, many outstanding studies have used state-of-art Artificial Intelligence (AI) techniques in the aim of diagnosing dementia through neuroimaging [4,5,6]. While it is not to be proved anymore the efficiency of such techniques, it remains hard to imagine the capability of constantly resorting to MRIs, especially on a large population scale. Other works have focused on cognitive clinical evaluation, and used feature selection for identifying a smaller subset of questionnaires items [7,8]. Yet, this still requires direct contact with the caregiver in the first place.
In the CAREPATH project, we take advantage of the Patient Empowerment Platform to drive ML research studies around “offline” early dementia detection. We develop indeed a set of new ideas and metrics that attempt to circumvent as much of direct clinical evaluations to the profit of monitored data through the CAREPATH facilities. While cognitive evaluation is constantly addressed through memory and logical games, we also leverage on movement, sleeping and mass data to evaluate the neuropsychiatric and daily activities components. We resort to features selection and clustering techniques in the context of unsupervised data. However, as the pilot goes on and labeled data become available through golden standard measurements, many other techniques can be used, such as SVMs, Naive Bayes, Random Forests and Neural Networks.
The exclusivity of our approach therefore resides in constantly evaluating individual’s dementia profile by means of ML algorithms, on the basis of device-monitored parameters that do not require any interaction with the caregiver.
Reference:
- Source: dementiastatistics.org/statistics/different-types-of-dementia/
- Staging Dementia Using Clinical Dementia Rating Scale Sum of Boxes Scores, A Texas Alzheimer's Research Consortium Study, Arch Neurol. 2008;65(8):1091-1095. doi:10.1001/archneur.65.8.1091, https://jamanetwork.com/journals/jamaneurology/fullarticle/796037
- Alzheimer’s Disease Cooperative Study ADL Scale, Jessica Fish, Encyclopedia of Clinical Neuropsychology pp 111–112, DOI: 10.1007/978-0-387-79948-3_1791, Springer
- Ewers, Michael, et al. "Neuroimaging markers for the prediction and early diagnosis of Alzheimer's disease dementia." Trends in neurosciences 34.8 (2011): 430-442
- Ahmed, Md Rishad, et al. "Neuroimaging and machine learning for dementia diagnosis: recent advancements and future prospects." IEEE reviews in biomedical engineering 12 (2018): 19-33.
- Frantellizzi, Viviana, et al. "Neuroimaging in vascular cognitive impairment and dementia: a systematic review." Journal of Alzheimer's Disease 73.4 (2020): 1279-1294.
- Kleiman, Michael J., et al. "Screening for early-stage Alzheimer’s disease using optimized feature sets and machine learning." Journal of Alzheimer's Disease 81.1 (2021): 355-366.
- Zhu, Fubao, et al. "Machine learning for the preliminary diagnosis of dementia." Scientific Programming 2020 (2020).