Pulmonary nodule segmentation and classification with explainable AI
Loading...
Date
2024-10-10
Authors
Στρατάκος, Κυριάκος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Lung cancer remains the leading cause of cancer-related mortality worldwide, making early and accurate detection critical. Pulmonary nodules, small growths in the lungs, are potential indicators of lung cancer and require precise classification to differentiate between benign and malignant cases. Over the past year, many techniques based on Machine Learning have been developed to address this issue. While achieving high accuracy, these techniques often fail to interpret and explain their classification decisions to physicians. In this thesis, we aim to address this issue by creating an explainable model for nodule classification. Using the LIDC-IDRI dataset, which contains annotated CT scans, we
developed a segmentation model to accurately identify and extract nodules. From these segmented volumes, we extracted radiomic features that capture various aspects such as texture, shape, and intensity. These features served as inputs to a Multi-Layer Perceptron (MLP) classifier, which was trained to distinguish between malignant and benign nodules. To enhance the interpretability of the model’s predictions, we employed the Local Interpretable Model-Agnostic Explanations (LIME) algorithm. LIME provided valuable insights into the decision-making process of the MLP, highlighting the most significant features influencing the classification outcomes. The proposed approach demonstrated high classification accuracy 76 % with sensitivity 85 %, with the explainability component ensuring that the model’s predictions could be trusted and understood by clinicians. This thesis contributes to the field of medical imaging and artificial intelligence (AI) by providing an interpretable method for pulmonary nodule classification, with potential applications in clinical practice.
Description
Keywords
Explainable AI, Machine learning, Artificial intelligence, Segmentation, Classification