EXACT-Net: Framework for EHR-Guided Lung Tumor Auto-Segmentation for Non-Small Cell Lung Cancer Radiotherapy

Area of Science:

• Artificial Intelligence in Medical Imaging
• Multimodal AI Frameworks
• Oncology

Background:

• Lung cancer, particularly non-small cell lung cancer (NSCLC), has a high mortality rate, with most patients requiring radiation therapy.
• Accurate tumor segmentation is crucial for timely NSCLC diagnosis and treatment, but manual methods are slow and prone to delays.
• Existing automated lung nodule detection methods, including deep learning models, often struggle with high false positive rates.

Purpose of the Study:

• To develop an automated lung tumor segmentation method that overcomes the limitations of existing approaches, specifically high false positive rates.
• To enhance the accuracy and efficiency of non-small cell lung cancer diagnosis and treatment planning.

Main Methods:

• Developed EXACT-Net (EHR-enhanced eXACtitude in Tumor segmentation), an AI model integrating electronic health records (EHRs) with imaging data.
• Utilized a pre-trained large language model (LLM) to extract relevant information from EHRs for improved tumor segmentation.
• Employed a zero-shot learning approach with the LLM to refine nodule detection and reduce false positives.

Main Results:

• The EXACT-Net model demonstrated a 250% increase in successful nodule detection.
• The approach effectively reduced false positives, improving the precision of automated lung tumor segmentation.
• The model was trained on computed tomography (CT) data from non-small cell lung cancer patients.

Conclusions:

• Combining vision and language information within the EXACT-Net multimodal AI framework significantly enhances performance over vision-only models.
• This study highlights the potential of multimodal AI in medical image processing for improved diagnostic accuracy.
• The EXACT-Net framework offers a promising pathway for more efficient and accurate lung cancer diagnosis and treatment.