Explainable AI for Interpretation of Ovarian Tumor Classification Using Enhanced ResNet50

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Deep learning models like ResNet and Inception show promise in classifying tumors.
• Current models lack explainability, hindering clinical adoption for decision-making.
• Identifying critical features for tumor classification is essential for healthcare practitioners.

Purpose of the Study:

• To implement a custom classifier for ovarian tumors with high performance.
• To interpret classification results using Explainable AI (XAI) methods.
• To identify image regions crucial for distinguishing benign from malignant ovarian tumors.

Main Methods:

• Utilized a dataset of ovarian tumor CT scans across multiple planes.
• Implemented a modified ResNet50 architecture, based on a pre-trained ResNet50.
• Applied various Explainable AI techniques for qualitative interpretation of model predictions.

Main Results:

• The modified ResNet50 achieved 97.5% classification accuracy on the test dataset.
• XAI analysis highlighted tumor shape and localized regions as significant features.
• These features are critical for determining metastatic potential and classifying tumors.

Conclusions:

• The developed model offers high accuracy and interpretability for ovarian tumor classification.
• Explainable AI successfully identified key visual features influencing malignancy classification.
• Findings support data-driven decisions and potential early detection of malignant ovarian tumors.