Comparison of CNN-Learned vs. Handcrafted Features for
Detection of Parkinson's Disease Dysgraphia in a Multilingual
Dataset

GALAZ, Zoltan, Peter DROTAR, Jiri MEKYSKA, Matej GAZDA, Jan MUCHA, Vojtech ZVONCAK, Zdenek SMEKAL, Marcos FAUNDEZ-ZANUY, Reinel CASTRILLON, Juan Rafael OROZCO-ARROYAVE, Steven RAPCSAK, Tamas KINCSES, Luboš BRABENEC and Irena REKTOROVÁ. Comparison of CNN-Learned vs. Handcrafted Features for Detection of Parkinson's Disease Dysgraphia in a Multilingual Dataset. Frontiers in Neuroinformatics. Lausanne: Frontiers Media SA, 2022, vol. 16, MAY, p. 877139-877156. ISSN 1662-5196. Available from: https://dx.doi.org/10.3389/fninf.2022.877139.

Other formats: BibTeX LaTeX RIS

TY  - JOUR
ID  - 2212066
AU  - Galaz, Zoltan - Drotar, Peter - Mekyska, Jiri - Gazda, Matej - Mucha, Jan - Zvoncak, Vojtech - Smekal, Zdenek - Faundez-Zanuy, Marcos - Castrillon, Reinel - Orozco-Arroyave, Juan Rafael - Rapcsak, Steven - Kincses, Tamas - Brabenec, Luboš - Rektorová, Irena
PY  - 2022
TI  - Comparison of CNN-Learned vs. Handcrafted Features for Detection of Parkinson's Disease Dysgraphia in a Multilingual Dataset
JF  - Frontiers in Neuroinformatics
VL  - 16
IS  - MAY
SP  - 877139
EP  - 877139
PB  - Frontiers Media SA
SN  - 16625196
KW  - machine learning
KW  - deep learning
KW  - feature extraction
KW  - Parkinson's disease dysgraphia
KW  - handwriting analysis
UR  - https://www.frontiersin.org/articles/10.3389/fninf.2022.877139/full
N2  - Parkinson's disease dysgraphia (PDYS), one of the earliest signs of Parkinson's disease (PD), has been researched as a promising biomarker of PD and as the target of a noninvasive and inexpensive approach to monitoring the progress of the disease. However, although several approaches to supportive PDYS diagnosis have been proposed (mainly based on handcrafted features (HF) extracted from online handwriting or the utilization of deep neural networks), it remains unclear which approach provides the highest discrimination power and how these approaches can be transferred between different datasets and languages. This study aims to compare classification performance based on two types of features: features automatically extracted by a pretrained convolutional neural network (CNN) and HF designed by human experts. Both approaches are evaluated on a multilingual dataset collected from 143 PD patients and 151 healthy controls in the Czech Republic, United States, Colombia, and Hungary. The subjects performed the spiral drawing task (SDT; a language-independent task) and the sentence writing task (SWT; a language-dependent task). Models based on logistic regression and gradient boosting were trained in several scenarios, specifically single language (SL), leave one language out (LOLO), and all languages combined (ALC). We found that the HF slightly outperformed the CNN-extracted features in all considered evaluation scenarios for the SWT. In detail, the following balanced accuracy (BACC) scores were achieved: SL-0.65 (HF), 0.58 (CNN); LOLO-0.65 (HF), 0.57 (CNN); and ALC-0.69 (HF), 0.66 (CNN). However, in the case of the SDT, features extracted by a CNN provided competitive results: SL-0.66 (HF), 0.62 (CNN); LOLO-0.56 (HF), 0.54 (CNN); and ALC-0.60 (HF), 0.60 (CNN). In summary, regarding the SWT, the HF outperformed the CNN-extracted features over 6% (mean BACC of 0.66 for HF, and 0.60 for CNN). In the case of the SDT, both feature sets provided almost identical classification performance (mean BACC of 0.60 for HF, and 0.58 for CNN).
ER  -

Basic information
Original name	Comparison of CNN-Learned vs. Handcrafted Features for Detection of Parkinson's Disease Dysgraphia in a Multilingual Dataset
Authors	GALAZ, Zoltan, Peter DROTAR, Jiri MEKYSKA, Matej GAZDA, Jan MUCHA, Vojtech ZVONCAK, Zdenek SMEKAL, Marcos FAUNDEZ-ZANUY, Reinel CASTRILLON, Juan Rafael OROZCO-ARROYAVE, Steven RAPCSAK, Tamas KINCSES, Luboš BRABENEC (203 Czech Republic, belonging to the institution) and Irena REKTOROVÁ (203 Czech Republic, belonging to the institution).
Edition	Frontiers in Neuroinformatics, Lausanne, Frontiers Media SA, 2022, 1662-5196.

Other information
Original language	English
Type of outcome	Article in a journal
Field of Study	30103 Neurosciences
Country of publisher	Switzerland
Confidentiality degree	is not subject to a state or trade secret
WWW	URL
Impact factor	Impact factor: 3.500
RIV identification code	RIV/00216224:14740/22:00129679
Organization unit	Central European Institute of Technology
Doi	http://dx.doi.org/10.3389/fninf.2022.877139
UT WoS	000811334000001
Keywords in English	machine learning; deep learning; feature extraction; Parkinson's disease dysgraphia; handwriting analysis
Tags	14110127, podil, rivok
Tags	International impact, Reviewed
Changed by	Changed by: Mgr. Pavla Foltynová, Ph.D., učo 106624. Changed: 19/8/2022 12:34.

Abstract

Parkinson's disease dysgraphia (PDYS), one of the earliest signs of Parkinson's disease (PD), has been researched as a promising biomarker of PD and as the target of a noninvasive and inexpensive approach to monitoring the progress of the disease. However, although several approaches to supportive PDYS diagnosis have been proposed (mainly based on handcrafted features (HF) extracted from online handwriting or the utilization of deep neural networks), it remains unclear which approach provides the highest discrimination power and how these approaches can be transferred between different datasets and languages. This study aims to compare classification performance based on two types of features: features automatically extracted by a pretrained convolutional neural network (CNN) and HF designed by human experts. Both approaches are evaluated on a multilingual dataset collected from 143 PD patients and 151 healthy controls in the Czech Republic, United States, Colombia, and Hungary. The subjects performed the spiral drawing task (SDT; a language-independent task) and the sentence writing task (SWT; a language-dependent task). Models based on logistic regression and gradient boosting were trained in several scenarios, specifically single language (SL), leave one language out (LOLO), and all languages combined (ALC). We found that the HF slightly outperformed the CNN-extracted features in all considered evaluation scenarios for the SWT. In detail, the following balanced accuracy (BACC) scores were achieved: SL-0.65 (HF), 0.58 (CNN); LOLO-0.65 (HF), 0.57 (CNN); and ALC-0.69 (HF), 0.66 (CNN). However, in the case of the SDT, features extracted by a CNN provided competitive results: SL-0.66 (HF), 0.62 (CNN); LOLO-0.56 (HF), 0.54 (CNN); and ALC-0.60 (HF), 0.60 (CNN). In summary, regarding the SWT, the HF outperformed the CNN-extracted features over 6% (mean BACC of 0.66 for HF, and 0.60 for CNN). In the case of the SDT, both feature sets provided almost identical classification performance (mean BACC of 0.60 for HF, and 0.58 for CNN).

Links
NU20-04-00294, research and development project	Name: Diagnostika onemocnění s Lewyho tělísky v prodromálním stadiu založená na analýze multimodálních dat
NU20-04-00294, research and development project	Investor: Ministry of Health of the CR, Diagnostics of Lewy body diseases in prodromal stage based on multimodal data analysis
734718, interní kód MU	Name: Novel Network-Based Approaches for Studying Cognitive Dysfunction in Behavioral Neurology (Acronym: CoBeN)
734718, interní kód MU	Investor: European Union, MSCA Marie Skłodowska-Curie Actions (Excellent Science)

PrintDisplayed: 8/5/2024 20:33

Comparison of CNN-Learned vs. Handcrafted Features for Detection of Parkinson's Disease Dysgraphia ...

Other applications