A Review on Parkinson’s Disease Detection Using Spiral Drawings and MRI Images

Abstract

Parkinson’s Disease (PD) is a progressive neurode generative disorder that affects millions globally, often going undiagnosed until motor symptoms become severe. Early and ac curate detection is essential for effective treatment and improved quality of life. In this study, we describe the use of Convolutional Neural Networks (CNNs) to recognize and classify Parkinson’s Disease using two key modalities: spiral drawings and MRI brain images. From 20 recent research papers, we present a com prehensive review of CNN-based approaches for PD detection, highlighting their advantages over traditional diagnostic methods. We introduce the basic concepts of CNNs and emphasize their ability to extract motor and structural features directly from raw images, outperforming manual scoring and conventional machine learning techniques.By combining motor and structural biomarkers, CNN-based systems offer a powerful tool for early PD detection. The fusion of spiral drawings and MRI images not only improves diagnostic accuracy but also supports neurologists in clinical decision-making. The use of AI in Parkinson’s Disease detection has the potential to transform healthcare delivery by enabling scalable, cost-effective, and patient-friendly solutions. CNN, ResNet-50, VGG-16, DenseNet, Spiral Drawing, MRI, Parkinson’s Disease, Deep Learning, Transfer Learning, Feder ated Learning, Mobile Diagnosis.

Introduction

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder affecting motor, cognitive, and emotional functions. Early diagnosis is crucial, but traditional clinical methods are subjective and often detect PD late. Recent research emphasizes non-invasive, data-driven approaches using motor tasks (e.g., spiral and wave drawings) and neuroimaging (MRI) to detect subtle signs of PD.

Motor Task Analysis: Spiral and wave drawings, when digitized, reveal tremors, pressure variations, and fine motor irregularities. Deep learning, particularly convolutional neural networks (CNNs), effectively extracts discriminative features from these drawings, often combined with geometric features (area, curvature, perimeter) for more robust and interpretable models. Transfer learning and data augmentation enhance accuracy and generalizability, while ensemble methods and hybrid architectures (CNN + SVM or dual CNNs) improve reliability. Traditional machine learning methods, like Random Forest or logistic regression using handcrafted kinematic features, remain competitive, especially for smaller datasets.

Neuroimaging Approaches: MRI-based PD detection focuses on structural and functional brain changes, particularly in the substantia nigra and basal ganglia. Deep learning models like DenseNet, ResNet, and hybrid CNN-LSTM architectures capture spatial and temporal patterns in MRI sequences. Lightweight models such as MobileNet and SqueezeNet enable mobile or remote deployment. Traditional methods, including SVMs with voxel-based features and explainable AI (e.g., SHAP analysis), provide interpretability, highlighting relevant brain regions.

Multimodal Fusion: Combining spiral/wave drawings with MRI data enhances diagnostic performance. Hybrid pipelines leverage both motor and structural indicators, yielding accuracies often exceeding 90%. Data preprocessing, augmentation, and feature fusion improve robustness, while lightweight networks support efficient clinical or mobile use.

Challenges and Future Directions: Key challenges include small datasets, variability in imaging or drawing protocols, and the “black-box” nature of deep learning models. Future research focuses on multimodal integration, explainable AI, generalizability across diverse datasets, and deployable systems for real-time screening and monitoring.

Conclusion

Recent advances confirm that deep learning models such as ResNet50, DenseNet, and VGG16, especially when combined with transfer learning, achieve high diagnostic accuracy for Parkinson’s disease from spiral test images, wave drawings and MRI scans often surpassing traditional machine learning techniques [3],[15]. These deep models automatically extract subtle motor or neuroimaging features vital for early diagnosis and have shown robust performance on research datasets [4]. Work on multimodal fusion demonstrates that combining spiral, wave and imaging data can provide a more holistic and sensitive screening tool for Parkinson’s [12],[16].Despite this notable progress, the deployment of such models in real world clinical environments poses important challenges. Many studies relied on relatively small, homogeneous datasets often limited to a single site or demographic profile raising concerns about generalizability across populations and disease stages [9],[18]. Furthermore, model interpretability remains a barrier to clinical acceptance; neurologists and clinicians require AI tools that offer transparent decision-making processes and ac tionable insights [7]. There is a consensus that explainable AI development, along with improved data curation and standard ization of spiral and MRI protocols will help bridge the gap from research to clinical practice [14],[16].The most promising future directions involve integrating deep learning frameworks with telemedicine solutions and wearable device ecosystems. Real time analysis of digitized spiral drawings or continuous motor signals, combined with remote expert supervision, may enable continuous monitoring, early detection and proactive disease management particularly benefiting patients in remote or underserved settings [13],[17]. Longitudinal and multi cen ter studies will be essential to validate the robustness and clinical value of these AI models in diverse settings [9],[20].In summary, deep learning approaches show clear advantages for automated detection and monitoring of Parkinson’s disease, yet their full clinical impact depends on solving open problems in data diversity, model transparency, and real world integration [14]

References

[1] Mrs. S. SasiRekha, Dr. R. Shankar, and Dr. S. Duraisamy, “Parkinsons Disease Prediction With Spiral Drawings and Wave Frequency Using Deep Conformal Neural Networks,” Journal of Neonatal Surgery, vol. 14, issue 4s, pp. 11–22, 2025 [2] V. V. Waykule, Siddhi Magdum, Abhishek Mule, Prachi Patil, and Mayank Ughade, “Detection of Parkinson’s Disease Using MRI and Spiral Images,” International Journal for Research in Applied Science and Engineering Technology, vol. 12, issue XI, November 2024, pp. 888-894. [3] Theyazn H. H. Aldhyani, Abdullah H. Al-Nefaie, and Deepika Koundal, “Modeling and Diagnosis Parkinson Disease by Using Hand Drawing: Deep Learning Model,” AIMS Mathematics, vol. 9, no. 3, pp. 6850–6877, 2024. DOI: 10.3934/math.2024334. [4] Megha Kamble, Prashant Shrivastava, and Megha Jain, “Digi tized Spiral Drawing Classification for Parkinson’s Disease Diag nosis,” Measurement: Sensors, vol. 16, pp. 100047, 2021. DOI: 10.1016/j.measen.2021.100047. [5] Priyal Agarwal, Vipin Talreja, Rutuja Patil, Vaishnavi Jadhav, and Indu Dokare, “Early Detection of Parkinson’s Disease Using Spiral Test,” in Data Science and Big Data Analytics, Springer, pp. 391–402, 2024. DOI: 10.1007/978-981-99-9179-2-30. [6] Nesren Farhah, “Utilizing Deep Learning Models in an Intelligent Spiral Drawing Classification System for Parkinson’s Disease Clas sification,” Frontiers in Medicine, vol. 11, pp. 1453743, 2024. DOI: 10.3389/fmed.2024.1453743. [7] Iman Beheshti and Ji Hyun Ko, “Predicting the Occurrence of Mild Cognitive Impairment in Parkinson’s Disease Using Structural MRI Data,” Frontiers in Neuroscience, vol. 18, pp. 1375395, 2024. DOI: 10.3389/fnins.2024.1375395. [8] Junyan Fu, Hongyi Chen, Chengling Xu, Zhongzheng Jia, Qingqing Lu, Haiyan Zhang, Yue Hu, Kun Lv, Jun Zhang, and Daoying Geng, “Harnessing Routine MRI for the Early Screening of Parkinson’s Disease: A Multicenter Machine Learning Study Using T2-Weighted FLAIR Imaging,” Insights into Imaging, vol. 16, no. 92, 2025. DOI: 10.1186/s13244-025-01961-3. [9] Nair Ul Islam, Ruqaiya Khanam, and Ashok Kumar, “Using 3D CNN for Classification of Parkinson’s Disease from Resting-State fMRI Data,” Journal of Engineering and Applied Science, vol. 70, no. 89, 2023. DOI: 10.1186/s44147-023-00236-2. [10] Hui Wen Loh, Wanrong Hong, Chui Ping Ooi, Subrata Chakraborty, Prabal Datta Barua, Ravinesh C. Deo, Jeffrey Soar, Elizabeth E. Palmer, and U. Rajendra Acharya, “Application of Deep Learning Models for Automated Identification of Parkinson’s Disease: A Review (2011–2021),” Sensors, vol. 21, no. 21, pp. 7034, 2021. DOI: 10.3390/s21217034. [11] Yingcong Huang, Kunal Chaturvedi, Al-Akhir Nayan, Mohammad Hesam Hesamian, Ali Braytee, and Mukesh Prasad, “Early Parkin son’s Disease Diagnosis through Hand-Drawn Spiral and Wave Anal ysis Using Deep Learning Techniques,” Information, vol. 15, no. 4, 220, 2024. DOI: 10.3390/info15040220. [12] Venkatesan Rajinikanth, Sahar Yassine, and Syed Ahmad Chan Bukhari, “Hand-Sketches based Parkinson’s disease Screening using Lightweight Deep-Learning with Two-Fold Training and Fused Opti mal Features,” International Journal of Mathematics, Statistics, and Computer Science, vol. 2, 2024. DOI: 10.59543/ijmscs.v2i.7821. [13] Nanziba Basnin, Nazmun Nahar, Fahmida Ahmed Anika, Mohammad Shahadat Hossain, and Karl Andersson, “Deep Learning Approach to Classify Parkinson’s Disease from MRI Samples,” Lecture Notes in Computer Science, pp. 536–547, 2021. DOI: 10.1007/978-3-030 86993-9-48. [14] Ramya R., Ramesh C., Murugesan P., Nithya N., and Sathish Kumar G., “Enhancing Parkinson’s Disease Prediction Using Deep Learning Based Convolutional Neural Networks,” J. Electrical Systems, vol. 20-5s, pp. 1866-1874, 2024. [15] Omar Alniemi and Hanaa F. Mahmood, “Convolutional neural net work for the detection of Parkinson disease based on hand-drawn spiral images,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 30, no. 1, pp. 267-275, April 2023. DOI: 10.11591/ijeecs.v30.i1.pp267-275. [16] Manuel Gil-Mart´ ?n, Juan Manuel Montero, and Rub´en San-Segundo, “Parkinson’s Disease Detection from Drawing Movements Using Convolutional Neural Networks,” Electronics, vol. 8, no. 8, 907, 2019. DOI: 10.3390/electronics8080907. [17] Sabyasachi Chakraborty, Satyabrata Aich, Jong-Seong-Sim, Eun young Han, Jinse Park, and Hee-Cheol Kim, “Parkinson’s Disease Detection from Spiral and Wave Drawings using Convolutional Neural Networks: A Multistage Classifier Approach,” in Proceedings of the 22nd International Conference on Advanced Communications Technology (ICACT), pp. 298–303, 2020. DOI: 10.23919/ICACT48636.2020.9061497. [18] Mohamad Alissa, Michael A. Lones, Jeremy Cosgrove, Jane E. Alty, Stuart Jamieson, Stephen L. Smith, and Marta Vallejo, “Parkinson’s disease diagnosis using convolutional neural networks and figure copying tasks,” Neural Computing and Applications, vol. 34, pp. 1433–1453, 2022. DOI: 10.1007/s00521-021-06469-7. [19] Meenakshi Malik, Edeh Michael Onyema, Mueen Uddin, Poonam Yadav, Aanchal Sharma, Jazlyn Jose, Achyut Shankar, Fahad Alasim, and Mustufa Haider Abidi, “Deep learning approach for Parkinson’s screening with geometric features from spiral and wave drawings,” Multimedia Tools and Applications, advance online publication, 2025. DOI: 10.1007/s11042-025-20915-x. [20] A. Anisha, Femima Shelly A. T, Benitta R. K, and Amala Selciya T. L., “Parkinson’s Disease Detection using Spiral Drawings,” InternatiIonal Journal of Innovative Science and Research Technology, vol. 8, issue 5, May–2023: 2658-2663, 2023

Copyright

Copyright © 2025 Aiswarya K B, Akhila A S, Arya Anish, Nahas V N, Ms. Krishna K B. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.