The HS-CMU Dataset for Diagnosing Benign and Malignant Diseases through Hysteroscopy

TL;DR

This paper introduces the HS-CMU dataset, comprising hysteroscopic videos and images annotated for diagnosing benign and malignant uterine diseases, aiming to support AI-driven diagnostic tools.

Contribution

The paper provides a new, annotated hysteroscopic image and video dataset to aid in differentiating benign and malignant intrauterine diseases.

Findings

Dataset includes 3385 high-quality images from 8 categories.

Videos recorded from 175 patients undergoing hysteroscopic surgery.

Annotations made by experienced clinicians to ensure accuracy.

Abstract

Hysteroscopy enables direct visualization of morphological changes in the endometrium, serving as an important means for screening, diagnosing, and treating intrauterine lesions. Accurate identification of the benign or malignant nature of diseases is crucial. However, the complexity and variability of uterine morphology increase the difficulty of identification, leading to missed diagnoses and misdiagnoses, often requiring the expertise of experienced gynecologists and pathologists. Here, we provide the video and image dataset of hysteroscopic examinations conducted at Beijing Chaoyang Hospital, Capital Medical University (named the HS-CMU dataset), recording videos of 175 patients undergoing hysteroscopic surgery to explore the uterine cavity. These data were obtained using corresponding supporting software. From these videos, 3385 high-quality images from 8 categories were selected to form the HS-CMU dataset. These images were annotated by two experienced obstetricians and gynecologists using lableme software. We hope that this dataset can be used as an auxiliary tool for the diagnosis of intrauterine benign and malignant diseases.