مشخصات پژوهش

عنوان	Toward Transformers for Accurate COVID-19 Detection using Chest X-Ray Images
نوع پژوهش	مقاله ارائه شده
کلیدواژه‌ها	Vision Transformers, Vits, Feature Extraction, Medical Images, Data Augmentation, Image Segmentation, Transformer Encoder
چکیده	This study presents a novel approach for feature extraction in medical images utilizing Vision Transformers (ViTs). Since ViTs benefit from diverse data, the method incorporates data augmentation techniques like rotation and zooming to enrich the dataset. The core approach involves segmenting the image into fixed-size patches, converting each patch into a vector representation, and incorporating positional information to capture spatial relationships. These enhanced patch representations are then fed into a standard transformer encoder for progressive extraction of high-level features from the medical image. This improved feature extraction has the potential to contribute to more accurate diagnoses. Furthermore, the proposed method achieves superior performance compared to various classification methods on a specific task. It achieves the highest scores across all metrics (precision, recall, F-score, and accuracy), surpassing a basic Support Vector Machine by over 1.3 in accuracy. These results suggest the promise of our method for medical image classification problems.
پژوهشگران	آرش خسروی (نفر دوم)، احمد آذرنیک (نفر سوم)، فائزه میرزایی (نفر اول)