Machine learning-based radiomic analysis and growth visualization for ablation site recurrence diagnosis in follow-up...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Detecting ablation site recurrence (ASR) after thermal ablation is challenging due to similar appearances of recurrence and post-ablative changes on CT scans.
• Radiomic analysis and machine learning offer potential solutions to improve ASR detection accuracy.

Purpose of the Study:

• To evaluate the efficacy of radiomic analysis in detecting ASR on follow-up computed tomography (CT) scans.
• To develop a visualization tool for emphasizing ASR between follow-up scans.

Main Methods:

• Radiomic features were extracted from regions of interest and modeled using Lasso regression and Extreme Gradient Boosting (XGBoost) classifiers.
• A leave-one-out test (LOOT) was used for performance evaluation.
• Difference heatmaps (diff-maps) were developed to visualize regions of growth indicative of ASR.

Main Results:

• Lasso regression with LOOT achieved an AUC of 0.97 and 92.73% accuracy.
• XGBoost achieved an AUC of 0.93 and 89.09% accuracy.
• Difference heatmaps accurately highlighted post-ablative liver tumor recurrence in all patients.

Conclusions:

• Machine learning-based radiomic analysis is effective for detecting ASR on follow-up CT scans.
• Growth visualization tools aid in highlighting potential ablation site recurrence.