Object Detection in Medical Imaging: A Quantitative and Qualitative Analysis of the Scientific Research (Preprint)

Abstract

Medical imaging is an indispensable tool widely employed in the healthcare industry and medical field. With the emergence of deep learning, medical imaging analysis has entered a new era where disease diagnosis and therapy assessment can be faster and less prone to error. Although there has been a significant increase in scientific research focusing on classification and segmentation tasks in medical imaging, the object detection task still deserves further exploration.

This study aims to evaluate and describe the latest advances and evidence regarding object detection algorithms in medical imaging tasks through quantitative and qualitative analysis, detecting research gaps and providing guidelines for future research.

We relied on Scopus and Web of Science to analyze the use of object detection algorithms in medical imaging, particularly on specific tasks and anatomical areas. The screening process and data extraction were conducted according to the PRISMA guidelines. We assessed the annual publication rate, research areas, preferred journals, leading research countries and authors, and selected keywords through quantitative analysis. Through qualitative analysis, we grouped articles with an annual citation rate higher than five based on the algorithm used, imaging modalities, and anatomical application areas.

In 2022, there was a significant increase in publications, accounting for 43% of all the identified articles. Most of these publications belong to the research fields of Computer Science, Medicine, and Engineering research fields. China, the United States of America, and Japan emerged as the most productive countries, with 124, 48, and 36 publications, respectively, out of the 311 identified articles. The Faster R-CNN algorithm is the most frequently used, with 32.5% of the publications, followed by the YOLO algorithm with 30%, and SSD with 8.75%. Common modalities include Computed Radiography, Pathological Imaging, Endoscopy, and Computed Tomography. Among the various application areas, digital pathology and microscopy were the most popular, with 16 publications, followed by the detection of abnormalities in abdominal organs, with 15 publications. Other areas such as abdominal organs focusing on colon polyp detection, digestive system organs emphasizing tooth-related tasks, bone imaging for fracture identification, and breast CT scans for breast lesion detection, were also frequently explored.

This study highlights the development and the increased popularity of object detection techniques in medical imaging for improving disease diagnosis and lesion detection. Additionally, it provides insights into potential collaboration and prospects for future research. The 80 most-cited articles were listed in greater detail, showing that object detection in medical imaging is a rapidly developing research area with potential for further growth.