Page 219 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 7 Early detection and diagnosis using deep learning 209

3.1 Rheumatic diseases
Rheumatic diseases including rheumatoid arthritis (RA), Sjög-
ren's syndrome, idiopathic inflammatory myositis, and the trio of
systemic sclerosis, lupus erythematosus, and vasculitis are
chronic autoimmune inflammatory diseases with involvement
of multiple organs. The unsteady interplay of many genetic and
environmental factors affects disease progression and develop-
ment. Rheumatic diseases are known as a single group, rather
than being identified as a single entity. This can be understood
keeping in mind the diverse and complex heterogeneity that
these diseases carry. Previous models for predicting risk for devel-
oping the disease and the results based on the scale of population
databases work well on average, but in terms of precision medi-
cine, many of the diagnosis and management needs of patients
with rheumatic diseases are still unmet. In this configuration,
DL may suggest effective solutions to the outstanding unresolved
problems resulting from such restless heterogeneous diseases.
The most obvious use of AI in a medical environment is
undoubtedly diagnostic assistance. This method uses a ML algo-
rithm, which requires a data set consisting of a collection of
clinical parameters and their diagnoses for a number of patients.
Using the parameters provided through the data set, the program
recognizes the relevant combination of variables required to pre-
dict an accurate diagnosis. These approaches to the problem are
helpful in the identification of biological signatures, highlighting
biomarkers of pathology. This approach is based on the resolu-
tion of classification problems; the training programs are called
“classifiers,” and they are trained to classify patients according
to their diagnosis.
After closely analyzing all the ML models currently working to
provide aid in this field, it can be seen that most of them usually
follow a two-step pattern. The first step is the detection of joints
by ML (using Haar cascading). The second step is a scoring of
joint destruction by DL (ConvNet), which comprises convolu-
tional layers, pooling layers, and fully connected layers. ConvNet
processes the input image as two-dimensional matrix data and
gives output as numerical values or probability of categories.
ConvNet is currently considered to be the most efficient algo-
rithm for image processing. The performance of the various
different models is determined by drawing lines of comparison
between the scores assigned by the models and the scores
assigned by rheumatologists that use radiographs, which were
not included in the validation process of the CNNs.