LNDb v4: pulmonary nodule annotation from medical reports

Area of Science:

• Medical Imaging
• Radiology
• Oncology

Background:

• Accurate lung cancer diagnosis is critical due to high prevalence.
• Radiologists identify pulmonary nodules through extensive radiology exam analysis.
• Computer-aided diagnosis (CAD) tools can enhance nodule identification efficiency.

Purpose of the Study:

• To convert unstructured medical reports into structured nodule annotations.
• To match textual information from reports with manually annotated data.
• To facilitate further research in lung cancer diagnosis and CAD.

Main Methods:

• Utilized medical reports and manually annotated data from the Lung Nodule Database (LNDb).
• Developed a method to extract nodule information from text and link it to image annotations.
• Created a tabular dataset of 292 medical reports with nodule characteristics and matches.

Main Results:

• Successfully converted 292 medical reports into nodule annotations.
• Generated files detailing nodule characteristics and their correspondence to manual annotations.
• Established a bridge between existing reports and manual annotations for research.

Conclusions:

• The created dataset supports research in lung nodule detection and computer-aided diagnosis.
• Enables comparative studies on the utility of text-derived versus manually curated annotations.
• Aims to advance lung cancer diagnostic accuracy and research methodologies.