Page 61 - AIH-1-2

Artificial Intelligence in Health AI in AD – Diagnosis and monitoring

Table 2. Overview of studies investigating diagnosis and monitoring of AD utilizing AI
Serial No. Applications Description References
1 CRM-PLS-DA Machine learning-assisted CRM along with PLS-DA aid in precise dermatological diagnosis of AD, 33
distinguishing AD from healthy individuals and exploring multiclass categorization of eczema severity.
2 CNN-based MPT An advanced learning system has been developed to diagnose AD using MPT images, 34
eliminating the need for manual intervention.
3 Multivariate machine learning The study utilized multivariate machine learning to create a diagnostic tool and severity 42
for AD severity prediction prediction model for AD patients, revealing intricate connections between clinical and serum
using combined biomarkers measures and enhancing disease understanding.
4. Machine learning classifier The study uses a machine learning classifier to accurately detect AD using gut epithelial 50
for AD detection using colonocytes and gut microbiota data. The robust pipeline includes techniques like feature
gut microbiota and selection, model selection, cross-validation, classification, and statistical assessments, enabling
transcriptome data precise discrimination based on omics data.
5 ANN for automated AD The ANN has been developed for automated diagnosis of AD. It uses a feed-forward 51
diagnosis architecture with nine input and output parameters, aiming to improve accuracy and
efficiency by distinguishing AD from other skin disorders and using its unique features for
classification purposes.
6 EczemaNet EczemaNet is a computer vision tool used to monitor AD severity. It uses CNN and transfer 52
learning techniques to predict severity. Clinical trials confirm its efficacy, providing a
standardized, effective monitoring approach.
7 Itch Tracker The Itch Tracker is a software application for smartwatches that tracks nocturnal scratching 71
and provides an objective assessment of itching. It uses an algorithm to analyze acceleration
data, distinguishing scratching from other movements based on unique wrist motions. The
device is effective in measuring pruritus severity in patients with AD.
8 Neurological imaging The application uses neurological imaging, specifically positron emission tomography and 75
functional magnetic resonance imaging, to objectively identify structural and functional
changes in pruritus, detecting brain activity during itching episodes.AI techniques could be
used in the analysis of neurological imaging data for pruritus research. These could include
automated image analysis, pattern recognition, and predictive modeling.
9 Acoustic surveillance Acoustic monitoring has been used to analyze scratching behavior in AD patients. Initially 67,68
applied to transgenic mice, a software application automates data analysis. A sound detector
integrated with wrist monitoring has accelerated data analysis, but more research is needed.
Abbreviations: AD: Atopic dermatitis; AI: Artificial intelligence; CNN: Convolutional neural network; MPT: Multiphoton tomography; CRM: Confocal
Raman microspectroscopy; PLS-DA: Partial least squares discriminant analysis; ANN: Artificial neural network.

which asks patients a single question: “On a scale of 0 – 10, genuine cases of pruritus. These cases demonstrate the
with 0 being ‘no itch’ and 10 being ‘worst itch imaginable,’ limitations of relying solely on scales and questionnaires.
how would you rate your itch at the worst point over While they may aid a doctor in managing an individual
the past 24 h?” The peak pruritus NRS has proven to be AD patient, more objective and supplementary tools are
a well-defined, reliable, sensitive, and accurate scale for desperately needed. 60-64 More specifically, rigorous research
determining the intensity of the most severe itching. Its in this area requires instruments capable of identifying
clear and simple format makes it particularly appealing to minute differences and facilitating comparisons among
busy clinicians. 59 cohorts before and after AD therapies or interventions. 65,66
Although scales and questionnaires have proven quite 7. Acoustic surveillance
useful in clinical settings, biases resulting from individual
differences in perceiving and expressing pruritus diminish Initially, acoustic surveillance was used on transgenic
their utility in research contexts. For example, what one mice specially engineered to mimic AD. The scratching
individual may rate as a “10 out of 10” or the “worst behavior of these animals was recorded using a sound
itch conceivable” might be noticeably less severe for recording device, and an in-depth analysis of the recorded
another, rendering such measures subjective. In addition, scratching noises, including the examination of frequency
not everyone experiences pruritus solely as an “itchy” and wavelength data, was conducted. Subsequently, a
sensation; some describe it as a “burning” or “tingling,” software application was developed to identify and measure
which could cause some clinical instruments to overlook the scratching habits of the mice. This method offers an

Volume 1 Issue 2 (2024) 55 doi: 10.36922/aih.2775