Specialized Image Descriptors Adaptation for Polyp Recognition over Endoscopic Images

This paper presents a novel approach to classification in biomedical imaging, specifically targeting polyp recognition in video endoscopy snapshots. Our method leverages specialized image descriptors to enhance the accuracy and robustness of polyp recognition. By employing these specialized descriptors, we address the challenges inherent in analyzing biomedical images from open datasets. Our approach not only improves classification performance but also offers a comparative efficiency compared to the existing heavyweight deep neural network model baseline. These results are confirmed by the following metric values: precision 0.983, recall 0.969, and F1-score 0.976. The proposed method demonstrates significant potential for advancing diagnostic capabilities in medical imaging and contributes to the ongoing efforts to develop more effective tools for automated medical analysis without significant computational resource requirements.