Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EXACT-Net:EHR-guided lung tumor auto-segmentation for non-small cell lung cancer radiotherapy

Hamed Hooshangnejad, Xue Feng, Gaofeng Huang, Rui Zhang, Katelyn Kelly, Quan Chen, Kai Ding

TL;DR

The paper addresses the challenge of timely, accurate lung tumor segmentation for NSCLC radiotherapy by reducing false positives through EHR-informed prior knowledge. It introduces EXACT-Net, an EHR-guided auto-segmentation framework that combines zero-shot LLM prompts to extract clinically confirmed tumor location with a 3D deep network (UNet3D/RetinaNet3D) to refine segmentation. On 10 unseen cases, the approach achieves a 70% ground-truth match and a 250% boost in successful nodule detection when EHR information is incorporated, highlighting the practical potential for accelerating treatment initiation. Limitations include false negatives in challenging cases and the need for broader validation, but the work demonstrates the value of integrating EHR-derived phenotype information with DL-based segmentation in NSCLC radiotherapy workflows.

Abstract

Lung cancer is a devastating disease with the highest mortality rate among cancer types. Over 60% of non-small cell lung cancer (NSCLC) patients, which accounts for 87% of diagnoses, require radiation therapy. Rapid treatment initiation significantly increases the patient's survival rate and reduces the mortality rate. Accurate tumor segmentation is a critical step in the diagnosis and treatment of NSCLC. Manual segmentation is time and labor-consuming and causes delays in treatment initiation. Although many lung nodule detection methods, including deep learning-based models, have been proposed, there is still a long-standing problem of high false positives (FPs) with most of these methods. Here, we developed an electronic health record (EHR) guided lung tumor auto-segmentation called EXACT-Net (EHR-enhanced eXACtitude in Tumor segmentation), where the extracted information from EHRs using a pre-trained large language model (LLM), was used to remove the FPs and keep the TP nodules only. The auto-segmentation model was trained on NSCLC patients' computed tomography (CT), and the pre-trained LLM was used with the zero-shot learning approach. Our approach resulted in a 250% boost in successful nodule detection using the data from ten NSCLC patients treated in our institution.

EXACT-Net:EHR-guided lung tumor auto-segmentation for non-small cell lung cancer radiotherapy

TL;DR

The paper addresses the challenge of timely, accurate lung tumor segmentation for NSCLC radiotherapy by reducing false positives through EHR-informed prior knowledge. It introduces EXACT-Net, an EHR-guided auto-segmentation framework that combines zero-shot LLM prompts to extract clinically confirmed tumor location with a 3D deep network (UNet3D/RetinaNet3D) to refine segmentation. On 10 unseen cases, the approach achieves a 70% ground-truth match and a 250% boost in successful nodule detection when EHR information is incorporated, highlighting the practical potential for accelerating treatment initiation. Limitations include false negatives in challenging cases and the need for broader validation, but the work demonstrates the value of integrating EHR-derived phenotype information with DL-based segmentation in NSCLC radiotherapy workflows.

Abstract

Lung cancer is a devastating disease with the highest mortality rate among cancer types. Over 60% of non-small cell lung cancer (NSCLC) patients, which accounts for 87% of diagnoses, require radiation therapy. Rapid treatment initiation significantly increases the patient's survival rate and reduces the mortality rate. Accurate tumor segmentation is a critical step in the diagnosis and treatment of NSCLC. Manual segmentation is time and labor-consuming and causes delays in treatment initiation. Although many lung nodule detection methods, including deep learning-based models, have been proposed, there is still a long-standing problem of high false positives (FPs) with most of these methods. Here, we developed an electronic health record (EHR) guided lung tumor auto-segmentation called EXACT-Net (EHR-enhanced eXACtitude in Tumor segmentation), where the extracted information from EHRs using a pre-trained large language model (LLM), was used to remove the FPs and keep the TP nodules only. The auto-segmentation model was trained on NSCLC patients' computed tomography (CT), and the pre-trained LLM was used with the zero-shot learning approach. Our approach resulted in a 250% boost in successful nodule detection using the data from ten NSCLC patients treated in our institution.
Paper Structure (18 sections, 4 equations, 6 figures, 3 tables)

This paper contains 18 sections, 4 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods and Materials
  4. LLM with Prompts for Tumor Phenotype Extraction
  5. Tumor Auto-segmentation Algorithm Design and Training
  6. Architecture
  7. Loss Functions
  8. Dataset and Data Preprocessing
  9. Data Augmentation
  10. Training and Evaluation Methods
  11. Experiment Design
  12. Results
  13. Tumor auto-segmentation performance
  14. LLM prompt design
  15. EHR-guided Tumor Auto-segmentation Experiment
  16. ...and 3 more sections

Figures (6)

  • Figure 1: The current radiotherapy workflow, consisting of multiple steps including the target and organs at risk segmenation.
  • Figure 2: Overview of EHR-guided automated target segmentation system. The auto-contouring platform contour the initial structures on the diagnostic CT scan. Due to advantages of PET scan for improved target segmentation, it will be used as the second primary imaging modality for target segmentation platform. The NLP based algorithm will extract critical information regarding the location and shape of tumor.
  • Figure 3: Detailed tumor auto-segmentation model architecture
  • Figure 4: Showing an example case of EHR-guided tumor auto-segmentation where the extracted information regarding the confirmed tumor is used to remove FPs. The nodule detection algorithm detected and classified seven nodules as malignant for this sample case (Case ID 4). Out of the seven nodules, three were found in right inferior lobe (RIL) and two were found in left inferior lobe (LIL) and two in left upper lobe (LUL). According to the EHR extracted information, however, the confirmed tumors were in LUL. Using this information, the result is automatically refined, and the FPs are removed.
  • Figure 5: Case 6 where the nodule is very close to chest wall making the nodule detection very challenging. Left showing the coronal view, and right axial view, with tumor marked with red circle.
  • ...and 1 more figures