Fig 1.
Schematic (a) and physical model (b) of the shape-sensing FBG system. 1- multipurpose handle; 2- three-lumen delivery catheter system; 3- Aurora EM sensor and wire; 4- bending control guidewire, 5 –bending control mechanism; 6 –the FBGS multicore shaper sensing fiber.
Fig 2.
The steps of the airway navigation process using the FBGS shape-sensing fiber.
1- Medial line computation; 2- Airway points (5 mm apart) placement along the median lines; 3- The software AIrShape generates a set of curves similar to the median lines to train the AI algorithm to identify the FBG catheter position; 4 –As the FBG catheter is advanced in the lung phantom, its shape is compared with the simulated curves and its position in a certain airway is identified; 5- AIrShape displays for median line position of the catheter for confirmation.
Fig 3.
CAD model of the airway phantom (light blue) with median (dark red) lines (a), and equally spaced (5 mm) markers along the median lines in the rectangle caption (b). Similar 5mm markers are placed on the median lines of all lung airways.
Fig 4.
Lung airways successfully navigated using the FBG catheter and the AIrShape software application (numbered 1–18).
Dashed line rectangles: upper lobe airways not navigated with this approach due to limitation of the bending radius of the FBG catheter.
Fig 5.
Test model to evaluate the FBG catheter’s flexibility, placed in 4 channels (a) and computed/acquired curves which approximate the catheter position (b): median line of channels (green), computed shape of the catheter using AIrShape software (red) and FBGS shape sensing system (blue). The red and blue curves fit on top of each other and deviate slightly from the green curve (median line) due to the rigid catheter’s tendency not to follow the median line.
Fig 6.
AIrShape MVCNN architecture and the complete navigation system data flow diagram from patient CT or lung phantom to median lines, AiRShape software and identification of airway trajectory to target (a), three orthogonal projections (views 1–3 above) computed for every median lines (b).
Fig 7.
The MVCNN architecture for identification of the median lines of lung airways.
Each median line is approximated with 500 AS curves and each AS curve is projected in three planes. For each projection, a CNN network is developed to identify its features. Individual CNN are fully connected into one output layer.
Table 1.
The parameters of the 2D CNN model for the three projections of each testing curve which approximates the position of the FBG catheter.
Fig 8.
AIrShape system setup for testing the shape sensing optical fiber in a custom-made delivery catheter.
Fig 9.
Output of the AIrShape software.
a: software display of the shape-sensing catheter position (green dot line) in the correct lung airway (red line) and the remaining airways (black lines). b: the position of the FBG catheter’s tip is confirmed with the iMTECH platform (red circle). The reference marker (grey circle), initial entry point (green circle) and the coordinate system (green arrows) are also shown.
Table 2.
The precision accuracy of identifying the correct lung airway by MVCNN during FBG navigation.
The results represent the mean and standard deviation of ten experiments.