Session Item

For MR-only treatment planning in skin High Dose Rate (HDR) surface applicator brachytherapy (SABT) it is crucial to detect applicator channels on MRI with high spatial accuracy. An optimized PETRA (Pointwise Encoding Time Reduction with Radial Acquisition) MR sequence has shown the potential to visualize Freiburg Flap (FF) applicators for skin SABT. This study aimed to develop an algorithm to automatically detect FF applicator catheters on PETRA images and estimate their positional accuracy by comparing calculated distances between them to their known distances in a FF.

An FF applicator (12 catheters, each passing through the center of 24 spheres with 10mm diameter) was fixed between two plexiglass blocks with its sides parallel to the block edges and placed on a flat MagPhan TMR008 phantom in a 3T Siemens Vida scanner (Fig1 a, b). Coronal and axial 3D PETRA images (TR\TE\TI 3.3\0.07\100 ms, FOV 280x280 mm², 0.8 mm isotropic voxels, BW 406 Hz/px) were acquired using a top UltraFlex Large 18 and bottom Spine 32 coil.

A custom MATLAB code selected a ROI comprising the FF in the coronal plane. The central column of the top left sphere was manually selected, and the central column position of top row reference spheres (RSs) to its right was estimated using pixel size. Row profiles of 13 columns corresponding to each RS were extracted and normalized and their numerical gradient was computed. Profile mean and standard deviation and gradient mean over columns were calculated for each RS (Fig 1c). Catheter positions were defined as row numbers corresponding to minima of profile mean over rows, found within 2 pixels of minima of standard deviation and of gradient mean. The distance between adjacent catheters was calculated from their row difference. Catheter distance difference from the known 10 mm value yielded positional accuracy. Average absolute positional accuracy and its standard deviation were calculated for all adjacent catheter pairs of each RS in a ROI (Fig 2c).

Catheters inside the FF were distinguished on PETRA images in all 3 orientations with higher signal intensity than air and lower signal intensity than silicone spheres (Fig 1 c; Fig 2 a, b). Automated catheter detection in the coronally placed applicator was feasible using the proposed algorithm for originally acquired and for reconstructed coronal slices centered on the applicator. The positional accuracy obtained was 0.40 ± 0.11 and 0.52 ± 0.30 for two acquired, and 0.46 ± 0.23 and 0.53 ± 0.30 (average ± standard deviation) for two reconstructed coronal slices encompassing the catheters.

The catheters of an FF applicator with known dimensions were detected on PETRA MR images using a novel algorithm. Overall catheter positional accuracy was around half millimeter, indicating reliable detection of catheters along their length inside the applicator. These results suggest that PETRA provides adequate catheter detection accuracy for application in SABT MR-only treatment planning.