European Radiology: Applications of radiomics in identifying Parkinson's disease

Parkinson's disease (PD) is one of the most common chronic progressive neurodegenerative diseases in the elderly population, caused by the early death of dopaminergic neurons in the central nervous system, which can produce a variety of motor and non-motor symptoms, seriously affecting the quality of life of
patients.
In addition, with the aging of the population, the annual incidence of PD continues to show an upward trend, which is expected to bring a huge economic and social burden
to society.
Therefore, early and accurate diagnosis of PD has become the key
to providing timely treatment and intervention to patients.

In recent years, molecular imaging techniques such as dopamine transporter (DAT) and fluorodeoxyglucose (FDG) PET imaging have played an increasingly important role
in the clinical diagnosis of PD.
Multiple studies have confirmed that normal DAT has high accuracy in excluding PD, but DAT imaging has some limitations
.
For example, due to the high cost of synthetic technology for DAT PET imaging, it is only implemented in a few medical centers, so it is not well popularized
.
In addition, its value in the differential and early diagnosis of atypical parkinsonism (APS) is limited.

] In contrast, many studies have shown that [18F]FDG PET brain imaging technology has shown good applicability in the early and differential diagnosis of neurodegenerative parkinsonism and has become an increasing research focus
.

Deep learning-based radiomics (DLR) is a newly developed method that can be used as an alternative to address the limitations of
radiomics.
The goal of the DLR method is to extract quantitative and high-throughput features from medical images through convolutional neural networks (CNNs), and then synergistically complement the image features with clinical information, thereby improving clinical decision-making capabilities
.
Since the features are extracted based on the deep learning network of the entire image, the DLR method does not require ROI segmentation or manual feature
extraction.

Recently, a study published in the journal European Radiology proposed a novel DLR method for computer-aided diagnosis of PD, and further clarified the value
of DLR characteristic foci extracted from [18F]FDG PET brain images in the diagnosis of PD and early aspects of PD.

This study recruited 255 normal control groups (NCs) and 103 patients with PD from Huashan Hospital in China.
26 NCs and 22 PD patients were recruited from Wuxi 904 Hospital in China as separate test groups
.
The proposed DLR model consists of a feature encoder based on a convolutional neural network and a classifier based on a support vector machine (SVM) model
.
The DLR model was trained and validated in the Huashan cohort and tested in the Wuxi cohort, with accuracy, sensitivity, specificity, and ROC curves to describe the performance of the model, and comparative experiments were performed in scale models, radiomics models, standard uptake ratio (SUVR) models, and DLR models.

Compared with other models, the DLR model showed advantages in differentiating patients with PD from NC, with an accuracy of 95.
17% [90.
35%, 98.
13%] (95% confidence interval, CI)
in the Huashan cohort.
In addition, the DLR model also showed stronger performance than conventional methods in early diagnosis of PD, with an accuracy of 85.
58% [78.
60%, 91.
57%]
in the Huashan cohort.

Figure ROC curve comparison
of four models in PD and NC classification.
(a) Comparison of PD and NC in the Huashan cohort, and (b) Comparison of PD and NC in the Wuxi cohort

In this study, a DLR method was developed to diagnose PD and PD subgroups, and the proposed model has significantly improved diagnostic performance compared with existing methods, providing a practical imaging method
for computer-aided diagnosis of PD.

Original source:

Xiaoming Sun,Jingjie Ge,Lanlan Li,et al.
Use of deep learning-based radiomics to differentiate Parkinson's disease patients from normal controls: a study based on [ 18 F]FDG PET imaging.
DOI:10.
1007/s00330-022-08799-z