A Deep Learning-Based Framework for Uncertainty Quantification in Medical Imaging Using the DropWeak Technique: An...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer remains a leading global cause of cancer mortality.
• Early detection significantly improves patient survival rates.
• Deep learning (DL) shows potential in medical diagnostics but requires rigorous accuracy evaluation.

Purpose of the Study:

• To evaluate the accuracy of various DL architectures for lung cancer classification.
• To incorporate uncertainty quantification into DL models for assessing classification confidence.
• To develop and validate a novel DL-based system for automatic lung cancer tumor classification from CT images.

Main Methods:

• Utilized multiple DL architectures, including Baresnet, for lung cancer classification tasks.
• Implemented uncertainty quantification techniques to measure the reliability of DL classification outputs.
• Developed an automatic tumor classification system using CT images.

Main Results:

• Achieved a classification accuracy of 97.19% for lung cancer detection using the developed DL system.
• Demonstrated the effectiveness of uncertainty quantification in improving the robustness of classification results.
• Validated the potential of DL with uncertainty analysis for accurate lung cancer diagnosis.

Conclusions:

• Deep learning, enhanced with uncertainty quantification, offers a promising approach for reliable lung cancer classification.
• The developed system can aid in achieving more accurate and dependable clinical diagnoses.
• Integrating uncertainty analysis is crucial for advancing DL applications in medical diagnostics.