Inventor(s): Yong Du, Kevin Leung, Martin Pomper, Steven Rowe
Unmet Need
Parkinson’s disease (PD) is the second most common neurodegenerative disorder, with approximately 60,000 Americans diagnosed with it each year. It is a progressive nervous system disorder that affects movement, from small tremors to a loss of automatic movements. One of the tools used for diagnosis of PD is Dopamine transporter (DAT) single photon emission computed tomography (SPECT) imaging. However, not only can the visual analysis of DAT-SPECT vary due to interobserver variability, but these DAT-SPECT images are also reconstructed iteratively and are strongly affected by reconstruction parameters. The images are also post processed heavily, often using smoothing to reduce noise, which can cause data loss. As there may be more information present in the projection data that is important for disease detection, there is a need for an improved method of using DAT-SPECT imaging to diagnose diseases like PD.
Technology Overview
Johns Hopkins researchers have developed a deep learning-based approach for detecting PD using raw projection data from DAT-SPECT imaging. A deep 3D convolutional neural network was developed, with 3D projection data used as input and the predicted likelihood that the subject was a healthy or PD patient as the output. It was trained on 527 patients, validated on 65 patients, and tested on 67 patients. In the test set, 21 patients were healthy and 46 had PD. Overall, the method yielded an accuracy of 97.0% on the test set. By using the DAT-SPECT 3D projection data as input to the neural network, the proposed method bypasses the need for image reconstruction, a time-consuming and variable process that causes data loss but was previously unavoidable in order to create human-interpretable images, and introduces a novel and automated approach to accurately detecting patients with PD.
Stage of Development
Proof of concept study.
Patent
N/A
Publication
K. H. Leung, W. Shao, L. Solnes, S. P. Rowe, M. G. Pomper, Y. Du. A deep-learning based approach for disease detection in the projection space of DAT-SPECT images of patients with Parkinson’s disease. J. Nucl. Med., 2020. 61 (supplement 1) 509.
