چاپ صفحه |  صفحه نخست سامانه |  نویسندگان |  اطلاعات تفضیلی |  دانلود مقاله | دانشگاه علوم پزشکی تبریز |
| نویسنده ثبت کننده مقاله | سحر رضائی |
| مرحله جاری مقاله | تایید نهایی |
| دانشکده/مرکز مربوطه | دانشکده پزشکی |
| کد مقاله | 87879 |
| عنوان فارسی مقاله | Deep Learning-Based Diagnosis and Risk Classification of Coronary Artery Disease in SPECT Myocardial Perfusion |
| عنوان لاتین مقاله | Deep Learning-Based Diagnosis and Risk Classification of Coronary Artery Disease in SPECT Myocardial Perfusion |
| نوع ارائه | پوستر |
| عنوان کنگره / همایش | 2025 SNMMI Annual Meeting |
| نوع کنگره / همایش | بین المللی |
| کشور محل برگزاری کنگره/ همایش | USA |
| شهر محل برگزاری کنگره/ همایش | New Orleans |
| سال انتشار/ ارائه شمسی | 1404 |
| سال انتشار/ارائه میلادی | 2025 |
| تاریخ شمسی شروع و خاتمه کنگره/همایش | 1404/03/31 الی 1404/04/03 |
| آدرس لینک مقاله/ همایش در شبکه اینترنت | https://snmmi.org |
| آدرس علمی (Affiliation) نویسنده متقاضی | Tabriz university of medical scinecses |
| نویسنده | نفر چندم مقاله |
|---|---|
| سحر رضائی | اول |
| اسماعیل قره پاپاق | دوم |
| مهسا کرباسی | سوم |
| عنوان | متن |
|---|---|
| خلاصه مقاله | Background: Coronary artery disease (CAD) remains one of the leading causes of morbidity and mortality worldwide. Myocardial Perfusion Imaging (MPI) using Single-Photon Emission Computed Tomography (SPECT) plays a crucial role in identifying ischemic regions and assessing disease severity. However, interpreting these images accurately demands a high level of clinical expertise and experience, which may not always be accessible. As artificial intelligence continues to advance, the application of deep learning models— particularly convolutional neural networks (CNNs)—offers a promising pathway to automate this diagnostic process. This study explores how transfer learning, using pretrained CNNs, can improve the classification of MPI SPECT images across different levels of disease severity, aiming to reduce diagnostic subjectivity and improve clinical decision-making Method: The research employed four pretrained CNN architectures: ResNet-50, ResNet-101, ResNet-152, and ShuffleNet V2. These models were fine-tuned on a custom MPI SPECT dataset to classify images under three distinct schemes: binary classification (e.g., Normal vs Severe), three-class classification (Normal, Mild, Intermediate), and four-class classification (Normal, Mild, Intermediate, Severe). The original output layers of each network were adjusted to match the number of target classes in each setup. Transfer learning allowed the models to leverage image features learned from large-scale datasets like ImageNet and adapt them to the more domain-specific medical imaging task. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics to capture both overall correctness and class-specific diagnostic effectiveness Results: In binary classification tasks, particularly distinguishing between normal and severely affected patients, the deeper ResNet architectures (ResNet- 101 and ResNet-152) achieved flawless performance with 100% accuracy. Meanwhile, ResNet-50 and ShuffleNet V2 reached 87.5%, indicating good, albeit slightly less robust, performance. When tasked with distinguishing mild abnormalities from normal conditions, ShuffleNet V2 performed best, achieving 95.6% accuracy. For intermediate severity classification, both ResNet-50 and ShuffleNet V2 excelled, achieving close to 98% accuracy. However, as the classification complexity increased to three and four classes, accuracy declined across all models. ResNet-152 led with 88.5% accuracy in the three-class task and 83.3% in the four-class task. The performance drop, particularly in four-class classification, highlighted challenges in handling imbalanced datasets and subtle visual differences between adjacent severity categories, especially for minority classes like the Severe group. Conclusion: Transfer learning with pretrained CNNs effectively improves automatic classification of MPI SPECT images, especially in clear-cut binary tasks. Although deeper networks like ResNet-152 perform slightly better, data quality and class balance are more critical. As classification complexity increases, performance declines due to subtle differences and imbalanced data. Future work should focus on balanced datasets and advanced techniques like data augmentation and class- weighted training to improve multi-class classification. Overall, transfer learning shows strong potential for reliable, automated CAD diagnosis. |
| کلمات کلیدی | Deep Learning, Risk Classification, Coronary Artery Disease, SPECT Myocardial Perfusion |
| نام فایل | تاریخ درج فایل | اندازه فایل | دانلود |
|---|---|---|---|
| SNMII_Poster_Template.pdf | 1404/04/06 | 208131 | دانلود |
| Gmail - 2025 SNMMI Annual Meeting Submission Notification - Abstract #252263.pdf | 1404/04/06 | 107479 | دانلود |
| conference_letter_C255064F-C484-349A-592C00F26DE6894A_050329_609.pdf | 1404/04/06 | 455509 | دانلود |
