A user-friendly deep learning application for accurate lung cancer diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate lung cancer diagnosis and treatment planning rely on clinician expertise.
• Deep learning in image processing offers faster, high-quality diagnostic tools.
• 3-D image processing from 2-D data faces limitations like organ superposition and distortion.

Purpose of the Study:

• To develop a radiomics and deep learning tool for lung cancer diagnosis.
• To enhance the accuracy and efficiency of lung cancer detection and classification.
• To support clinicians in analyzing medical images for precise treatment planning.

Main Methods:

• Applied deep learning (U-NET for segmentation, DenseNet for classification) to 1098 lung CT scans (86 from Bach Mai Hospital, 1012 open-source).
• Utilized radiomics to measure lung nodule characteristics like diameter, surface area, and volume.
• Developed a web-based interface using Python and Streamlit, integrated with Arduino and MFRC522 for data management.

Main Results:

• The segmentation model achieved a validation loss of 0.498 and training loss of 0.27.
• The cancer classification model demonstrated a validation loss of 0.78 and a training accuracy of 0.98.
• The system successfully recognized and classified lung cancer from chest CT scans.

Conclusions:

• The developed tool successfully aids in lung cancer diagnosis and classification from CT scans.
• The system facilitates direct storage and updating of patient data, improving accessibility for healthcare providers.
• The integrated system enhances clinical communication, information exchange, and cancer diagnosis summaries.