saad02/brain-tumor-single-slice-MRI-scan-with-synthetic-ehr-africa

该数据集捆绑了单切片脑部MRI扫描和丰富的结构化合成电子健康记录（EHR）数据，适用于多模态机器学习研究。每行包含一个`image`结构（字节+路径）、一个源`label`列和一个EHR负载，其中包含完整的JSON记录和方便的摘要列。合成的EHR数据聚焦非洲地区：它们编码了国家、城乡、设施级别、保险、就诊时间、常见合并症（如HIV/TB/疟疾）、现实程序（CPT/SNOMED/ICD-10-PCS）、诊断（ICD-10）、带剂量和ATC代码的药物、纵向遭遇和结果时间线。临床规则和严格的验证器确保核心一致性（例如，替莫唑胺仅在肿瘤病例中使用；癫痫发作时使用AEDs；农村/初级设置中的延迟更长）。所有数据都是合成的且非识别性的。

This dataset bundles single-slice brain MRI scans and richly structured, synthetic EHR data into a single Parquet file suitable for multimodal ML research. Each row contains an `image` struct (bytes + path), a source `label` column, and an EHR payload with both a full JSON record and convenient summary columns. The synthetic EHRs are Africa-focused: they encode country, urban/rural, facility level, insurance, time-to-presentation, common comorbidities (e.g., HIV/TB/malaria), realistic procedures (CPT/SNOMED/ICD-10-PCS), diagnoses (ICD-10), medications with dosing and ATC codes, longitudinal encounters, and outcomes timelines. Clinical rules and a strict validator enforce core coherence (e.g., temozolomide only in tumor cases; AEDs when seizures are present; longer delays for rural/primary settings). All data are synthetic and non-identifiable.