AI-powered PET imaging analysis for enhanced Parkinson’s disease diagnosis

سال انتشار: 1402
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 88

نسخه کامل این مقاله ارائه نشده است و در دسترس نمی باشد

این مقاله در بخشهای موضوعی زیر دسته بندی شده است:

استخراج به نرم افزارهای پژوهشی:

لینک ثابت به این مقاله:

شناسه ملی سند علمی:

IBIS12_200

تاریخ نمایه سازی: 12 آبان 1403

چکیده مقاله:

Parkinson’s disease (PD), a prevalent neurodegenerative disorder characterized by striataldopamine deficiency, poses challenges in early and accurate diagnosis, often leading tomisclassification with atypical parkinsonism. While conventional imaging techniques like PET andSPECT detect changes in striatal dopamine levels and functional changes and correlate them with motorresponses, they necessitate expert interpretation. This study addresses these diagnostic complexities byleveraging machine learning (ML) and convolutional neural networks (CNN), to automate and expeditePD accurate diagnostics. Our research focuses on analyzing ۴۰۰ PET images from visits at months ۱۲,۳۶, and ۴۸ from the Parkinson’s Progression Markers Initiative (PPMI) dataset. The goal is to identifysimilarities between brain images of PD patients for enhanced diagnostic accuracy. In the first step, weresize our ۳D images (۱۲۸x۱۲۸x۱۲۸) to ۲D images (۱۲۸x۱۲۸) by applying average pooling to reducecomplexity. Subsequently, we further resize these images to ۲D (۶۴x۶۴). Following the preprocessingsteps, we employ a convolutional autoencoder to extract similar features, thereby enhancing theclustering operation. After clustering using MeanShift, we analyze each cluster for every pair of originalimages. For each pair, we calculate the correlation matrix and average all correlation matrices to obtaina single correlation matrix. In the last step, we scrutinize the final correlation matrix to identify cellswith values closer to one. We map these cells to the corresponding pixels, and by aggregating highcorrelationpixels, we identify the regions that are repeated in our images. Employing advanced AIalgorithms, our methodology trains models to recognize recurring similar regions across images. Thisenables the detection and computation of specific brain regions, potentially indicative of PD. Automatedand rapid diagnosis by integrating ML and CNN into PET imaging analysis has the potential to enhancepersonalized treatment strategies and assist doctors in facilitating the diagnostic process.

نویسندگان

A Fallah

School of Mathematics, Statistics and Computer Science, University of Tehran, Tehran, Iran

Y Zandiyevakili

School of Mathematics, Statistics and Computer Science, University of Tehran, Tehran, Iran

Z Salehi

Hematology, Oncology and Stem Cell Transplantation Research Center, Tehran University of Medical Sciences, Tehran, Iran- Research Institute for Oncology, Hematology and Cell Therapy, Tehran University of Medical Sciences, Tehran, Iran

K Kavousi

Institute of Biochemistry and Biophysics (IBB), Department of Bioinformatics, Laboratory of Complex Biological Systems and Bioinformatics (CBB), University of Tehran, Tehran, Iran