Signs of Parkinson’s disease seem to be all around us. Scientists have discovered indicators of the disease’s onset in hair, blood, and even earwax. Speech patterns, mental health, and even place of residence are all linked to varying degrees to this condition, which affects movement and muscle control.
However, despite the potential of these signals, Parkinson’s disease—considered the world’s fastest-growing neurological disorder in terms of disability and mortality—still relies heavily on clinical examinations for formal diagnosis. The lack of readily available, reliable biomarkers also hinders early diagnosis.
Yet, a testing concept described in the journal Discover Computing may point toward a new path forward.
In a recent study, a team of scientists from India developed an AI-based system capable of detecting Parkinson’s disease with remarkable accuracy using the results of a simple drawing test.
Analyzing handwriting and drawings is viewed as a promising non-invasive screening tool for future disease detection; scientists are increasingly combining this approach with AI tools capable of identifying signs of neurodegenerative diseases based on subtle nuances in handwriting.
In an experiment led by computer scientist Ishan Ayus from Siksha ‘O’ Anusandhan University, researchers utilized data from a previous study on Parkinson’s disease conducted in Brazil involving 66 individuals.
This dataset included 31 people with Parkinson’s disease and 35 healthy controls; participants performed drawing exercises using a biometric “smart pen” that captured their drawings while simultaneously recording signals related to their hand movements. During the experiments, participants traced two types of symbols: spiral shapes and meanders (patterns consisting of continuous angular lines).
Ayus and his colleagues then fed the images and sensor data recorded by the smart pen into a series of different deep learning systems.
Each model evaluated spatial irregularities in the drawn spiral and meander patterns and analyzed subtle differences in motor skills and hand movement coordination captured by the pen.
The data was then processed using the SNAKE algorithm, which factored in and weighted the various AI model assessments before ultimately determining whether each drawing was created by a person with Parkinson’s disease or a healthy control subject.
Self-injection device
Makes injections comfortable once and for all. The Spasilen injector is simple and reliable!
According to the researchers, their multimodal system outperforms several other AI systems previously used to detect Parkinson’s disease via handwriting and drawings, correctly identifying meander patterns with 98.95% accuracy and spiral patterns with 97.74% accuracy.
The team attributes these results to the capabilities of the SNAKE algorithm, which balances predictions from various deep learning systems using an accuracy-weighted model, as well as to the dual measurements provided by the smart pens.
“Handwritten images provide spatial characteristics regarding stroke irregularities, tremor-induced distortions, and shape deviations,” the researchers write. “In contrast, handwriting signals captured by sensors record temporal changes in motor activity, including fluctuations in speed, uneven pressure, and coordination deficits.”
The researchers acknowledge that the small dataset used (comprising just 66 individuals) does not accurately reflect the clinical diversity of Parkinson’s disease across larger populations; they state that their findings should be viewed as an indication of their method’s potential rather than definitive proof of its diagnostic efficacy.
Nevertheless, this establishes a promising new benchmark for detecting Parkinson’s disease using such methods; if validated in larger-scale clinical trials, the team believes their approach could play a vital role in simplifying the diagnosis of the disease.
“Overall, the study demonstrates the growing role of AI in the early and accessible diagnosis of neurological disorders, supporting the ongoing shift toward preventive and personalized medicine,” the researchers write. “The proposed concept could ultimately facilitate non-invasive, low-cost, and remotely accessible neurological screening and decision support of clinical decisions.”
About the Author
