Radiolytx-DOPA

Description

Radiolytx-DOPA is designed specifically for the early diagnosis of PD in the sub-clinical case.

The workflow can be summarized as follow:

A) Unstructured data expressed in Natural Language during the patients anamnesis include a description of eventual motion impairment, unexpected depression, sudden sleep disorders, impairment of olfactive capabilities and subjective statements of the medical doctor. A series of key-words are extracted from the anamnesis text.

B) Structured data expressed as Numbers from a series of 630 metabolomic quantities.

C) A dataset with 100 PD patients with early diagnosis and 100 negative controls is available for this project, thanks to the collaboration with the University of Molise and NEUROMED.
The keywords are interpreted as binary values (yes/no). The 630 parameters are first reduced with a Principal Components Analysis. The principal components expressing 99.5% of the total variance are selected. A Convolutional Neural Network (CNN) with inputs the key-words and the selected principal components is used as a classifier of patients with risk of PD.

D) If NO Risk for PD is detected, the medical doctor can take as granted the result for further diagnosis.

E) If a risk for PD is detected, [18F]-DOPA PET scan is recommended. PET images are unstructured data, from which quantitative information needs to be extracted. At a subclinical stage the result of a [18F]-DOPA PET scan may be negative, but dedicated textural analysis in the regions mainly affected by PD will show significant deviations from healthy subjects. 103 textural parameters are extracted from the 89 Hammers brain regions, resulting in additional 8779 parameters. For each brain region, the parameters are reduced with Principal Components Analysis, selecting only these principal components expressing 99.5% of the total variance. Convolutional Neural Network (CNN) with inputs the key-words and the selected principal components from the metabolomics and the PET data is trained to classify the PD risk.