Auxiliary diagnosis model for pathological classification of cervical cancer based on biomarkers in radiomics

Mei Wang^1,2Yu Cao³Mengchen Zhu³Peilin Zhao⁴Qin Zhou⁵Hongxiang Lan⁶Jizhao Liu⁶Junqiang Lei^1,7,8*

• ¹First Hospital of Lanzhou University, Lanzhou, China
• ²Department of Obstetrics and Gynecology, The First Hospital of Lanzhou University, Lanzhou, China
• ³The First Clinical Medical College, Lanzhou University, Lanzhou, China
• ⁴The Second Clinical Medical College, Lanzhou University, Lanzhou, China
• ⁵School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, China
• ⁶School of Information Science and Engineering, Lanzhou University, Lanzhou, China
• ⁷Gansu Province Clinical Research Renter for Radiology Imaging, The First Hospital of Lanzhou University, Lanzhou, China
• ⁸Intelligent Imaging Medical Engineering Research Center of Gansu Province, The First Hospital of Lanzhou University, Lanzhou, China

Radiomics can provide objective and accurate decision support for disease diagnosis, prognosis evaluation and individualized treatment by extracting high-throughput image features and constructing prediction models. However, in existing cervical cancer radiomics research, lesion-area segmentation still depends on manual division. In addition, the characteristics of cervical cancer lesions are complex and changeable, which makes it difficult to extract meaningful imaging features and construct effective biomarkers, resulting in difficulties in the clinical diagnosis and screening of cervical cancer. In this work , based on the anatomical and imaging characteristics of cervical cancer, we designed a deep learning lesion area extraction method for cervical cancer images and designed a histogram feature that reflects the pixel concentration trend of the lesion area, and integrated it with existing radiomics features and clinical features to perform feature engineering screening, forming a cervical cancer pathological classification diagnosis model based on radiomics biomarkers. The results show that our method can effectively extract the lesion features of cervical cancer MRI images. Further, we found that 30 imaging features such as Median of Histogram, GLSZM Large Area Low Gray Level Emphasis (LoG, σ = 2.0mm, 3D), GLRLM Long Run Low Gray Level Emphasis (LoG, σ = 2.0mm, 3D) are important biomarkers for the diagnosis of cervical cancer. The auxiliary diagnosis model of cervical cancer pathological classification based on radiomics biomarkers constructed in this study provides an objective and effective auxiliary diagnosis tool for clinicians. It performs well in the prediction of cervical cancer pathological classification, and shows the potential to reduce unnecessary blood and Sample et al. Running Title histological examination, which is helpful for the improvement of early screening and clinical diagnosis methods of cervical cancer.