基于乳腺旋切术的超声肿瘤智能检测方法研究

Research on Intelligent Ultrasound Tumor Detection Method Based on Breast Rotary Resection

摘要: 乳腺癌是全球女性中最常见的恶性肿瘤之一，且其发病率呈逐年上升趋势，给全球公共卫生体系带来了巨大压力。随着医学影像技术的进步，尤其是超声成像技术的提升，超声引导下的乳腺肿瘤微创旋切术逐渐成为一种重要的治疗手段，因其创伤小、恢复快而被广泛应用。然而，现有手术方法仍依赖外科医生的经验，导致肿瘤定位精度不足，尤其在手术过程中肿瘤的实时定位和切除仍面临较大挑战。为解决这一问题，本研究提出了一种基于乳腺旋切术的超声肿瘤智能检测模型，结合深度学习和图像处理算法，旨在显著提高肿瘤检测的准确性。本文收集167例患者手术视频并抽取4020张乳腺肿瘤的超声图像，并使用YOLOv8和Faster R-CNN两种深度学习模型进行训练与比较，评估其在检测速度和精度方面的表现。实验结果表明，YOLOv8在检测速度上显著优于Faster R-CNN，且其精确率达到0.9902，明显高于Faster R-CNN的0.9694。该模型能够有效辅助外科医生进行精准的肿瘤定位和切除，提升了术中的准确性和效率，具有广泛的推广前景和临床应用潜力。

Abstract: Breast cancer is one of the most common malignant tumors among women worldwide, and its incidence rate is increasing year by year, exerting tremendous pressure on the global public health system. With the advancement of medical imaging technology, especially the improvement of ultrasound imaging technology, minimally invasive rotary resection of breast tumors under ultrasound guidance has gradually become an important treatment method, and is widely used due to its small trauma and quick recovery. However, the current surgical methods still rely on the experience of surgeons, resulting in insufficient tumor localization accuracy. Especially during the surgical process, the real-time localization and resection of tumors still face significant challenges. To address this issue, this study proposes an intelligent ultrasound tumor detection model based on breast rotary resection, combining deep learning and image processing algorithms, aiming to significantly enhance the accuracy of tumor detection. In this paper, surgical videos of 167 patients were collected and 4020 ultrasound images of breast tumors were extracted. Two deep learning models, YOLOv8 and Faster R-CNN, were used for training and comparison to evaluate their performance in terms of detection speed and accuracy. The experimental results show that YOLOv8 is significantly superior to Faster R-CNN in detection speed, and its accuracy rate reaches 0.9902, which is significantly higher than 0.9694 of Faster R-CNN. This model can effectively assist surgeons in precise tumor location and resection, improving intraoperative accuracy and efficiency, and has broad promotion prospects and clinical application potential.