Skip to main content

Table 6 Selected highest ranking AI feature differences between Parkinson disease and controls (Mann–Whitney U test), and between stages (Kruskal–Wallis test)

From: Antemortem detection of Parkinson’s disease pathology in peripheral biopsies using artificial intelligence

AI features

Mann–Whitney U test

U

p Value

PreciseDx Graph Feature27

143.0

0.001

LTS cluster size StDev

122.0

0.001

PreciseDx Graph Feature3

138.0

0.002

PreciseDx Graph9

170.0

0.009

StDev of Hematoxylin Channel

175.5

0.025

PreciseDx Graph Feature14

210.0

0.027

Calinski-Harabasz index

179.5

0.030

PreciseDx Graph Feature13

201.5

0.047

PreciseDx Graph Feature28

208.5

0.056

Davies-Bouldin Index

204.0

0.088

PreciseDx Graph Feature11

249.0

0.338

PreciseDx Graph Feature10

268.0

0.419

C index

266.5

0.601

LTS pixel fraction

168.0

0.017

StDev of DAB channel (brown)

173.5

0.023

Number of LTS patches

153.0

0.008

 

Kruskal–Wallis test

 

Kruskal–Wallis H

p Value

PreciseDx Graph Feature4

11.6

0.003

PreciseDx Graph Feature17

10.9

0.004

PreciseDx Graph Feature3

7.5

0.024

LTS cluster size StDev

7.0

0.03

PreciseDx Graph Feature27

4.8

0.089

  1. Statistically significant values are shown in bold
  2. Controls, n = 14; PD patients, n = 42; PD early stage patients, n = 15; PD moderate stage patients, n = 13; PD advanced stage patients, n = 14
  3. LTS Lewy type synucleinopathy