This repository contains the code for An Ensemble Multi-View 3D Convolution Neural Network Model for Lung Adenocarcinoma Risk Stratification on Thin Slice Computed Tomography: A Multi-Center Study.
For a comprehensive guide on using our web-based system at seeyourlung.com.cn, please refer to our tutorial video below. For a higher resolution version of the video, please visit Bilibili.
seeyourlung-tutorial.MP4
Lung cancer is among the most frequently diagnosed cancers worldwide. However, few studies predict the invasive grades of lung adenocarcinoma, an important task that can assist in planning a suitable surgical approach (lobectomy or sublobar resection) prior to operation.
We propose an ensemble multi-view 3D convolution neural network (EMV-3D-CNN) model to comprehensively study the risk grades of lung adenocarcinoma. Our codes consist of three main parts: preprocessing, training, and evaluation.
Figure 1 shows the flowchart of the proposed EMV-3D-CNN model. It involves three key tasks: diagnosing benign and malignant lung tumors (Task 1), classifying between pre-invasive and invasive lung tumors (Task 2), and identifying the risk stratification (i.e., Grade 1, Grade 2, Grade 3) of invasive lung tumors (Task 3).
We provide the trained models for various 3D medical image analysis tasks, which can be downloaded from BaiduYun(verification code: 3acd).
For easier access, we have also implemented the proposed model as a web-based system at seeyourlung.com.cn. By uploading the full original CT images in DICOM format, our algorithm can assign risk grades to pulmonary nodules given the center location of the target lung nodule.
The code is written in Python and requires the following packages:
- Python 3.9.12
- TensorFlow 2.8.0
- Keras 2.8.9
- Matplotlib 3.5.2
- Numpy 1.22.4
- Pandas 1.4.2
- Sklearn 1.1.1
- Scipy 1.8.1
- Install Python 3.9.12
- pip install -r requirements.txt