Parkinson's disease development prediction by c-granule computing compared to different AI methods

• Published in: Journal of information and telecommunication (Print) 2020-10, Vol.4 (4), p.425-439
• Main Authors: Przybyszewski, Andrzej W., Śledzianowski, Albert
• Format: Article
• Language: English
• Subjects: Accuracy; aggregation; Computation; disease model; disease progression; Fuzzy set theory; Fuzzy sets; Granular computing; Granular materials; Machine learning; Parkinson's disease; Set theory; Signs and symptoms; similarity
• ISSN: 2475-1839; 2475-1847
• DOI: 10.1080/24751839.2020.1749410

Both rough set theory (RST) and fuzzy rough set theory (FRST) are related to intelligent granular computing (GrC) primarily with the help of static granules. Our granules are sets of attributes measured from Parkinson's disease (PD) patient in a certain moment of his/her disease. Our complex granule (c-granule) approach was used to model longitudinal PD development. With RST/FRST we were looking for similarities between attributes of patients in different disease stages to more advanced PD patients. We have compared group (G1) of 23 PD with attributes measured three times (visits V1-V3) every half of the year (G1V1, G1V2, G1V3) to the other group of 24 more advanced PD (G2V1). By means of RST/FRST, we have found rules describing symptoms of G2V1 and applied them to G1V1, G1V2, and G1V3. With RST (FRST), we've got the following accuracies: G1V1 −59 (38)%; G1V2 - 68 (54)%; G1V3 - 86 (61)%, but global coverage for FRST was better. We also tried to compare results with several different machine learning methods, obtaining accuracies of G1V1 - 59%, G1V2 - 73%, and G1V3 - 78%. In summary, several different methods confirmed that generally one group of PD patients during disease development become more similar to a different group of more advanced PD.