Uterine disorders classification based on symptomic, behaviorial data and ultrasound images

The Uterine Disorder Diagnostic Multimodal Dataset (UDDMD-2025) consists of 223 anonymized patient records collected through structured interviews and clinical ultrasound images and reports from the Department of Gynecology, R.S. Unit, Shaheed Tajuddin Ahmad Medical College Hospital, Bangladesh. Data collection took place between May 5–August 7, 2025 (95 days). Each participant record includes diagnostic parameters extracted from ultrasound and medical reports, along with symptomatic and behavioral indicators gathered in person. All participants provided informed consent and shared their experiences without external influence. The text dataset is provided as a single .xlsx file (also convertible to .csv), where each row represents one participant and each column corresponds to clinical or behavioral variables. The dataset is publicly accessible via Mendeley Data (DOI: 10.17632/pdzpdgc497.1). To support multimodal research, the dataset includes a complete set of raw ultrasound images. For every participant in the .xlsx file, a corresponding image directory is provided, named P1, P2, …, P223, allowing direct mapping between tabular data and image evidence. Images were captured using Google Pixel 6 and iPhone 15 Pro devices and are shared in their original, unedited form for flexible preprocessing and analysis. Ethical clearance was obtained from the Institutional Ethics Committee of Daffodil International University, and permission for clinical data collection was granted by the Assistant Director of Shaheed Tajuddin Ahmad Medical College Hospital. All diagnostic fields were verified by departmental experts, and ultrasound summaries were manually entered and checked for consistency. The .xlsx file and image folders contain no personally identifiable information, ensuring full anonymization. Ultrasound summaries were collected immediately after physician consultations, ensuring the reliability of diagnostic information. The combination of raw images and structured text allows robust multimodal applications, including image–text alignment, diagnostic modeling, and cross-modal learning. This dataset supports research in: -Predictive modeling of uterine disorders -Behavioral–clinical correlation analysis -Machine learning for reproductive health -Multimodal medical data integration -Image–text paired training for medical AI By integrating validated diagnostics, raw ultrasound images, and behavioral indicators, UDDMD-2025 offers a strong foundation for advancing AI-based women’s health research and evidence-driven gynecological analytics. The authors acknowledge Md Wakil Ahmed and Rokonozzaman Ayon for their assistance in data transcription and digitization during the data preparation phase of this study.子宫疾病诊断多模态数据集（UDDMD-2025）包含223条经过匿名化处理的患者记录，数据采集自孟加拉国沙希德·塔吉丁·艾哈迈德医科大学附属医院妇科R.S.科室的结构化访谈记录、临床超声影像及报告。数据采集时间为2025年5月5日至8月7日，共计95天。 每份受试者记录均包含从超声影像与医疗报告中提取的诊断参数，以及现场收集的症状与行为指标。所有受试者均已签署知情同意书，且在不受外部干扰的情况下分享了自身相关经历。 文本数据集以单个.xlsx文件形式提供（亦可转换为.csv格式），其中每一行代表一名受试者，每一列对应一项临床或行为变量。该数据集可通过Mendeley Data公开获取（DOI：10.17632/pdzpdgc497.1）。 为支持多模态研究，数据集还包含完整的原始超声影像集。针对.xlsx文件中的每一位受试者，均提供了对应的影像目录，命名格式为P1、P2……P223，可实现表格数据与影像证据的直接映射。影像采用Google Pixel 6与iPhone 15 Pro设备拍摄，以原始未编辑的形式共享，以便灵活开展预处理与分析工作。 本研究已获得达弗德国际大学机构伦理委员会的伦理批准，同时获取了沙希德·塔吉丁·艾哈迈德医科大学附属医院助理主任的临床数据采集许可。 所有诊断字段均经科室专家核验，超声总结报告由人工录入并核查一致性。.xlsx文件与影像文件夹均未包含任何个人可识别信息，实现了完全匿名化处理。 超声总结报告于医师会诊后即刻收集，确保了诊断信息的可靠性。原始影像与结构化文本的结合可支撑丰富的多模态研究应用，包括影像-文本对齐、诊断建模及跨模态学习。 本数据集可支持以下方向的研究： - 子宫疾病预测建模 - 行为-临床相关性分析 - 生殖健康领域机器学习应用 - 多模态医疗数据集成 - 医疗AI的影像-文本配对训练 通过整合经过验证的诊断数据、原始超声影像与行为指标，UDDMD-2025为推动基于人工智能的女性健康研究及循证妇科分析提供了坚实的研究基础。 作者感谢Md Wakil Ahmed与Rokonozzaman Ayon在本研究的数据准备阶段，为数据转录与数字化工作提供的协助。