=======================

How AI can help in Alzheimer’s Disease?

For the last few days, i was posted in Psychiatry posting. To make things more interesting, i  decided to do projects related to whichever posting i attend. While i was very interested in creating interacting platform of psychiatry patients, 15 days isn’t enough time for that!

After stumbling on several ideas, i finally came across a problem of Diagnosis of Alzheimer’s disease. This blog is about my journey about figuring out the right questions to answer!

What is Alzheimer’s Disease?

Alzheimer disease (AD) is a neurodegenerative disorder of uncertain cause and pathogenesis that primarily affects older adults and is the most common cause of dementia.

It occurs due to accumulation and deposition of cerebral amyloid-β (Aβ) and is the most common cerebral amyloid deposition disease.

The most essential and often earliest clinical manifestation of AD is selective memory impairment, although there are exceptions. While treatments are available that can ameliorate some symptoms of the illness, there is no cure or disease-modifying therapy (treatment that slows the course of the illness) currently available, and the disease inevitably progresses in all patients.

Epidemiology- Alzheimer disease is the most common cause of dementia, responsible for 60-80% of all dementias. The prevalence is strongly linked to age, with >1% of 60-64-year-old patients being diagnosed with the condition, compared to 20-40% of those over 85-90 years of age 2.

Risk factors include

• advanced age
• female gender
• apolipoprotein E (ApoE) ε4 (epsilon 4) allele carrier status
• current smoking
• family history of dementia
• Down syndrome

In addition to the genetic and environmental factors above, the age of presentation is also influenced by socioeconomic factors:

• formal education
• income
• occupational status
• social network and family support

Mild Cognitive Impairment or the Prodromal Phase of AD

The prodromal phase of AD is commonly referred to as MCI, which is characterized by the onset of cognitive symptoms (e.g, memory or other cognitive dysfunctions) that do not meet the criteria for dementia.

The time course of AD is approximately 20–30 years from preclinical (cognitive normal) to prodromal (mild cognitive impairment [MCI] or pre-dementia) to overt AD.

Prevalence – The prevalence of MCI in non-dementia individuals over the age of 70 years is approximately 15% with a 2:1 ratio of amnestic to non-amnestic types

Diagnosis of MCI  – The National Institute on Aging-Alzheimer’s Association developed a new diagnostic criteria for AD, MCI due to AD, and preclinical AD that integrate biomarker evidence into the diagnostic frame.

In these criteria, concomitant observation of amyloid deposition (PET or CSF) and neuronal injury (tau, FDG-PET, MRI) indicates a high likelihood of MCI due to AD, while the presence of only one of these factors indicates an intermediate likelihood of MCI due to AD.

Recent International Working Group-2 criteria further simplified the diagnosis of AD based on the requirement of only the presence of an appropriate clinical phenotype at any stage and presence of amyloid biomarker (e.g., decreased Aβ in CSF, or increased tracer retention on amyloid PET).

Future predictions about the count of Alzheimer’s – “According to the 2014 World Alzheimer’s Report, dementia affects approximately 44 million people worldwide, and the incidence of Alzheimer’s disease (AD) is expected to triple by the year 2050 (1)”

Etiology – Previous studies have indicated that approximately 10–15% of MCI patients progress to AD yearly, and predictors of this conversion include the APOE4 allele of the apolipoprotein E gene (APOE), clinical severity, brain atrophy, certain cerebrospinal fluid (CSF) biomarkers, changes in cerebral glucose metabolism, and Aβ deposition.

Genetics – APOE4 allele of the apolipoprotein E gene (APOE) Pathologically – Alzheimer’s disease is characterized by the accumulation of two abnormal proteins: extracellular Aβ protein and intracellular tau protein.

Patients with PSEN1 mutations had the earliest median age of onset (43 years). The range of symptom onset across all mutation types is nonetheless fairly broad, with some presentations in the fourth decade and some mutations not manifesting symptoms until the seventh decade. Individuals with Down syndrome, who have an additional gene dose of APP due to trisomy of chromosome 21, inevitably develop AD pathology, and symptoms emerge at an earlier age, 10 to 20 years younger than the general population with AD

Biochemically – AD begins with the abnormal metabolism of the transmembrane amyloid precursor protein

(APP). β- and γ-secretases cleave APPs to form several Aβ peptide fragments. Of these, the most important is Aβ42, which is highly prone to aggregation and resultant plaque formation. Although amyloid deposits are typically observed in the extracellular space, Aβ is also found within neurons, and this may be related to the aggregation of other cellular proteins such as tau protein in AD.

Pathology – Amyloid and tau deposition progress spatiotemporally in a predictive manner. Amyloid first accumulates in the basal part of the frontal, temporal, and occipital lobes, and subsequently spreads to the entorhinal cortex, hippocampus, amygdala, insular cortex, and cingulate cortex, sparing the primary visual and sensorimotor cortices. Conversely, neurofibrillary tangle deposition progresses in the following order: transentorhinal cortex, entorhinal cortex, hippocampus, temporal cortex, association cortices, and finally the primary sensorimotor and visual cortices.

The underlying reason for this accumulation is poorly understood, as does the reason for non-uniform distribution in the cortex. There is, however, increasing evidence to suggest that chronic inflammation is at least partially responsible. Such inflammation can lead to prolonged parenchymal activation of microglial cells which in turn results in the release of inflammatory mediators with subsequent neuronal damage and amyloid-induced neurodegeneration.

Radiologically – The imaging of dementia has not only diagnostic and prognostic potential, but may also grant unique insights into dementia itself.

Cortical Atrophy in AD and Other Dementias Accumulation of amyloid plaques and neurofibrillary tangles in AD is contemplated to induce neural and synaptic loss that finally leads to cortical atrophy. Cortical atrophy typically occurs first in the hippocampus and associated entorhinal cortex prior to pervasive progression, and is considered to be an early marker of neuro degeneration.

CT – demonstrates the characteristic patterns of cortical atrophy,

MRI – But MRI is more sensitive to these changes and better able to exclude other causes of dementia (e.g. multi-infarct dementia) and as such is the favored modality.

The primary role of MRI (and CT for that matter) in the diagnosis of Alzheimer disease is the assessment of volume change in characteristic locations which can yield a diagnostic accuracy of up to 87%. Unfortunately, such volume loss is not apparent early in the course of the disease. The diagnosis should be made on the basis of two features:

1. mesial temporal lobe atrophy (particularly the hippocampus, entorhinal cortex and perirhinal cortex)
2. temporoparietal cortical atrophy

(source - Mina Park et. al)

Mesial temporal lobe atrophy can be assessed directly or indirectly.

Direct assessment is of hippocampal or parahippocampal volume loss while indirect assessment relies on an enlargement of the parahippocampal fissures. The former is more sensitive and specific but ideally, requires actual volumetric calculations rather than ‘eyeballing’ the scan. These measures have been combined in the medial temporal atrophy score which has been shown to be predictive of progression from mild cognitive impairment (MCI) to dementia.

In posterior cortical atrophy or early-onset Alzheimer disease, presents with parietal atrophy. This is often best seen on the interhemispheric surface of the parietal lobe by examining the posterior cingulate sulcal and parieto-occipital sulcal size and degree of atrophy of the precuneus and cortical surface of the parietal lobe. This has also been combined into a scoring system.

Brain volume measurements, assessed with segmentation, demonstrate that patients with Alzheimer disease have accelerated rates of brain volume loss, typically around twice normal (1% vs ~0.5% per year). This is even more marked in the hippocampi, with affected individuals exhibiting three times the volume loss per year (~4.5% vs ~1.5% per year). Nuclear medicine SPECT and PET are able to detect regional hypoperfusion/hypometabolism in a biparietal and bitemporal distribution.

FDG PET F-18 fluorodeoxyglucose (FDG) PET typically shows bilateral temporoparietal, precuneus and posterior cingulate hypometabolism which is usually symmetric. Uptake may be asymmetric in the early stages. The anterior cingulate, visual cortex (provided the eyes were kept open during uptake time), basal ganglia, thalami, occipital lobes and cerebellum are usually spared. Classically, even late in the disease, the sensorimotor cortices are relatively spared. Frontal lobes may be involved in late stage. With increased cerebral amyloid-β (Aβ) deposition, increased activity is demonstrated in the cortex.

It is particularly useful in excluding Alzheimer disease as the cause of dementia, as a negative amyloid PET scan renders the diagnosis unlikely.

(source - Mina Park et. al)

Diffuse Cortical Atrophy – Diffuse cortical atrophy is a common finding in various pathological states such as stroke, radiation treatment, and neurodegenerative disorders as well as in normal aging.

The global cortical atrophy (GCA) scale was first introduced as a tool to quantify this kind of atrophy in stroke patients with or without dementia The GCA evaluates atrophy in 13 different brain regions (frontal, parieto-occipital, and temporal sulcal dilation, and dilatation of the ventricles) and assigns a subscore (0 to 3) at each of these levels.

As an alternative, a four-point ventricular enlargement scale (0 to 3) was developed to quantify enlargement of the lateral ventricles in an axial orientation, and this scale has also demonstrated very good inter-observer agreement (intra-class correlation coefficient = 0.85).

Medial Temporal Lobar Atrophy

(source - Mina Park et. al)

Recently, the NINCDS-ADRDA suggested incorporation of structural MRI of the medial temporal lobe into the criteria for probable AD, but this suggestion did not specify a method for atrophy evaluation.

Posterior Cortical Atrophy

Early posterior cerebral involvement has emerged as an important aspect of AD, and appears to be a feature of early-onset rather than late-onset AD. Therefore, posterior cortical atrophy (PCA) combined with relative sparing of the medial temporal lobe may characterize atypical presentations of AD patients.

(source - Mina Park et. al)

Clinical features –

Age of onset – AD is characteristically a disease of older age. It is exceptional for AD to occur before age 60. The incidence and prevalence of AD increase exponentially with age, essentially doubling in prevalence every 5 years after the age of 65 years.

Individuals with premorbid higher function/supports are able to compensate for early disease changes to a greater degree and thus present later. Consequently, when well-supported patients eventually present, they tend to have more marked morphological changes on imaging.

Early-onset AD (onset of symptoms before 65 years of age) is unusual, and many of these patients present for evaluation due to concerns about job performance. These account for less than 1 percent of all cases of AD. They typically exhibit an autosomal dominant inheritance pattern related to mutations in genes that alter beta-amyloid (Aβ) protein production or metabolism, including amyloid precursor protein (APP), presenilin-1 (PSEN1), and presenilin-2 (PSEN2).

Clinically characterized predominantly by memory deficits. There are number of atypical clinical patterns exist, which are nonetheless pathologically Alzheimer disease. The typical patient presents initially with antegrade episodic memory deficits.

Clinically features arrive well before they are identified to be pathologically identical to Alzheimer disease, are characterized by slowly progressive focal cortical atrophy, with symptoms and signs matched to the affected area. Examples include:

• posterior cortical atrophy
• frontal variant of Alzheimer disease
• minority of cases with predominant semantic dementia

Diagnostic protocols

Diagnosis Clinical diagnosis is made by identifying a progressive decline in memory both with clinical examinations and neuropsychologic tests and has been historically based on the NINCDS-ADRA criteria, which divides patients according to the certainty of the diagnosis into:

1. definite: clinical diagnosis and histologic confirmation
2. probable: typical clinical syndrome without histologic confirmation (81% sensitive, 73% specific)
3. possible: atypical clinical features without histologic confirmation but no alternative diagnosis

Although using longitudinal clinical criteria is highly sensitive in diagnosing a dementia of any type (>90%), they are relatively inaccurate (<70%) in diagnosing Alzheimer disease specifically.

Importantly, the NINCDS-ADRA criteria only include imaging and laboratory examination or blood and CSF in excluding other causes. The only definitive diagnostic test is brain biopsy which in practice is rarely obtained.

(source - Mina Park et. al)

Treatment and prognosis – There is no cure for this disease; some drugs have been developed trying to improve symptoms or, at least, temporarily slow down their progression.

• cholinesterase inhibitors e.g. donepezil
• partial NMDA receptor antagonists
• medications for behavioral symptoms
• antidepressants
• anxiolytics
• antiparkinsonian (movement symptoms)
• anticonvulsants/sedatives (behavioral)

Differential diagnosis – The main differential is limbic-predominant age-related TDP-43 encephalopathy (LATE) – Bilateral mesial temporal lobe volume loss on imaging in an elderly amnestic patient. Pathologically distinct, but it can also co-exist with Alzheimer’s disease. At present, there is no definitive way of establishing the diagnosis other than retrospectively at autopsy. Some ways to distinguish

• rostrocaudal involvement of the amygdala and hippocampus (amygdala involved first followed by hippocampal head and anterior body)
• profound and asymmetric amygdala and hippocampal volume loss with relatively little involvement of other parts of the brain.
• negative amyloid-PET

My Goals for Project

After all the above readings, it seems convincing that there are significant changes  occuring in Alzheimer’s disease. During next few days i aim to achieve following goals –

• AI tool to detect Alzheimer’s Disease from MRI scans
• Classification of Alzheimer’s Disease into different clincal stages
• Explainability of model, for radiologists to check
• Prognostic marking of patient in his current state

In next few days, i would release the project that i worked upon and what all findings i was able to extract. So, stay tunned.

Resources used in this blog –

Structural MR Imaging in the Diagnosis of Alzheimer’s -Disease and Other Neurodegenerative Dementia:Current Imaging Approach and Future Perspectives – Mina Park, MD, Won-Jin Moon, MD, PhD                    https://doi.org/10.3348/kjr.2016.17.6.827 pISSN 1229-6929 · eISSN 2005-8330              Korean J Radiol 2016;17(6):827-845

https://www.uptodate.com/contents/clinical-features-and-diagnosis-of-alzheimer-disease

https://www.mayoclinic.org/diseases-conditions/alzheimers-disease/symptoms-causes/syc-20350447

https://radiopaedia.org/articles/alzheimer-disease-1