Assisted Probe Positioning for Ultrasound Guided Radiotherapy Using Image Sequence Classification

Effective transperineal ultrasound image guidance in prostate external beam\nradiotherapy requires consistent alignment between probe and prostate at each\nsession during patient set-up. Probe placement and ultrasound image\ninter-pretation are manual tasks contingent upon operator skill, leading to\ninteroperator uncertainties that degrade radiotherapy precision. We demonstrate\na method for ensuring accurate probe placement through joint classification of\nimages and probe position data. Using a multi-input multi-task algorithm,\nspatial coordinate data from an optically tracked ultrasound probe is combined\nwith an image clas-sifier using a recurrent neural network to generate two sets\nof predictions in real-time. The first set identifies relevant prostate anatomy\nvisible in the field of view using the classes: outside prostate, prostate\nperiphery, prostate centre. The second set recommends a probe angular\nadjustment to achieve alignment between the probe and prostate centre with the\nclasses: move left, move right, stop. The algo-rithm was trained and tested on\n9,743 clinical images from 61 treatment sessions across 32 patients. We\nevaluated classification accuracy against class labels de-rived from three\nexperienced observers at 2/3 and 3/3 agreement thresholds. For images with\nunanimous consensus between observers, anatomical classification accuracy was\n97.2% and probe adjustment accuracy was 94.9%. The algorithm identified optimal\nprobe alignment within a mean (standard deviation) range of 3.7$^{\\circ}$\n(1.2$^{\\circ}$) from angle labels with full observer consensus, comparable to\nthe 2.8$^{\\circ}$ (2.6$^{\\circ}$) mean interobserver range. We propose such an\nalgorithm could assist ra-diotherapy practitioners with limited experience of\nultrasound image interpreta-tion by providing effective real-time feedback\nduring patient set-up.\n