Ensembled liver cancer detection and classification using CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Computed tomography (CT) is crucial for diagnosing liver disease, but tumor characterization can be challenging for radiologists due to image intensity variations.
• Computer-assisted diagnostic techniques are increasingly important for improving the accuracy and efficiency of liver tumor diagnosis.

Purpose of the Study:

• To develop and evaluate a computer-assisted system for detecting liver tumors in CT images.
• To classify detected liver tumors into Hepato Cellular Carcinoma (HCC) and Metastases (MET).
• To enhance diagnostic performance using a multi-level ensemble model.

Main Methods:

• Evaluation of six different classifiers for tumor detection and classification from liver CT images.
• Development of a multi-level ensemble model integrating features extracted from CT scans.
• Utilizing k-fold cross-validation (CV) to ensure the robustness and reliability of the classification models.

Main Results:

• Individual classifiers achieved high accuracy for tumor identification (98.39%–100%) and moderate accuracy for tumor classification (76.38%–87.01%).
• The multi-level ensemble model demonstrated superior performance in both liver tumor detection and classification compared to individual classifiers.
• The developed system provides automated tumor characterization, aiding radiologists in early-stage diagnosis.

Conclusions:

• Automated tumor characterization using AI on liver CT images can significantly assist radiologists.
• The multi-level ensemble model offers a robust and accurate approach for detecting and classifying liver tumors (HCC vs. MET).
• This technology has the potential to improve early detection rates and patient outcomes for liver cancer.