Classification of Pulmonary Nodular Findings based on Characterization of Change using Radiology Reports.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Radiology reports require manual classification of findings for clinical action.
• Manual classification is time-consuming, error-prone, and critical for patient care.
• Automating the characterization of pulmonary nodular findings is a significant challenge.

Purpose of the Study:

• To develop and evaluate a framework for automated detection and classification of pulmonary nodular findings in radiology reports.
• To improve the efficiency and accuracy of identifying clinically significant changes in lung nodules.
• To address the limitations of manual data analysis in radiological workflows.

Main Methods:

• A framework combining a pre-trained word embedding model and a deep learning sentence encoder was developed.
• A Siamese network with pairwise inputs was utilized to overcome limited labeled training data.
• A multitask neural network classifier was designed for robust classification of nodule characteristics.

Main Results:

• The proposed framework demonstrated promising performance in classifying pulmonary nodular findings.
• The system effectively detects and characterizes changes in lung nodules from radiology reports.
• Performance was evaluated against state-of-the-art approaches, showing competitive results.

Conclusions:

• The developed framework offers an effective solution for automating the classification of pulmonary nodular findings.
• This approach can enhance the radiology workflow by reducing manual effort and potential errors.
• The study highlights the potential of deep learning and Siamese networks in medical text analysis for improved patient management.