Detecting and Characterizing Inferior Vena Cava Filters on Abdominal Computed Tomography with Data-Driven...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Inferior Vena Cava (IVC) filters are crucial for preventing pulmonary embolism but can lead to complications if not managed properly.
• Accurate detection and characterization of IVC filters on abdominal computed tomography (CT) scans are essential for clinical decision-making.
• Existing methods for IVC filter assessment can be time-consuming and prone to human error.

Purpose of the Study:

• To develop and evaluate data-driven algorithms for automated detection and characterization of IVC filters on abdominal CT scans.
• To assist healthcare providers in managing IVC filter placement and removal, thereby reducing associated complications.
• To establish a computational framework for identifying patients with retained IVC filters for individualized treatment decisions.

Main Methods:

• Development of two algorithms: a 2D + transfer learning (2D + TL) model and a 3D + recurrent convolutional neural network (3D + RCNN) model.
• Training computational models using Tensorflow Keras API on a dataset of 2048 abdominal CT studies from 439 patients.
• Utilizing a fine-tuning supervised training strategy with reference annotations for filter location and type.

Main Results:

• The 3D + RCNN model achieved higher sensitivity (0.923) and precision (0.853) for filter detection compared to the 2D + TL model (0.911 sensitivity, 0.804 precision).
• Both models demonstrated high confidence in predicting IVC filter locations (0.993 for 2D + TL, 0.996 for 3D + RCNN).
• The filter type prediction component achieved high performance, with the 3D + RCNN model showing 0.940 sensitivity, 0.927 specificity, and 0.975 AUC.

Conclusions:

• The developed AI framework shows significant promise in accurately detecting and characterizing IVC filters from abdominal CT scans.
• These algorithms can serve as valuable tools for healthcare providers, aiding in the timely and appropriate management of IVC filters.
• This study represents a novel contribution to automated IVC filter analysis, paving the way for improved patient outcomes and reduced medical complications.