Comparison of the impact of skull density ratio with alternative skull metrics on magnetic resonance-guided focused ultrasound thalamotomy for tremor.

OBJECTIVE

One of the key metrics that is used to predict the likelihood of success of MR-guided focused ultrasound (MRgFUS) thalamotomy is the overall calvarial skull density ratio (SDR). However, this measure does not fully predict the sonication parameters that would be required or the technical success rates. The authors aimed to assess other skull characteristics that may also contribute to technical success.

METHODS

The authors retrospectively studied consecutive patients with essential tremor who were treated by MRgFUS at their center between 2017 and 2021. They evaluated the correlation between the different treatment parameters, particularly maximum power and energy delivered, with a range of patients' skull metrics and demographics. Machine learning algorithms were applied to investigate whether sonication parameters could be predicted from skull density metrics alone and whether including combined local transducer SDRs with overall calvarial SDR would increase model accuracy.

RESULTS

A total of 62 patients were included in the study. The mean age was 77.1 (SD 9.2) years, and 78% of treatments (49/63) were performed in males. The mean SDR was 0.51 (SD 0.10). Among the evaluated metrics, SDR had the highest correlation with the maximum power used in treatment (ρ = -0.626, p < 0.001; proportion of local SDR values ≤ 0.8 group also had ρ = +0.626, p < 0.001) and maximum energy delivered (ρ = -0.680, p < 0.001). Machine learning algorithms achieved a moderate ability to predict maximum power and energy required from the local and overall SDRs (accuracy of approximately 80% for maximum power and approximately 55% for maximum energy), and high ability to predict average maximum temperature reached from the local and overall SDRs (approximately 95% accuracy).

CONCLUSIONS

The authors compared a number of skull metrics against SDR and showed that SDR was one of the best indicators of treatment parameters when used alone. In addition, a number of other machine learning algorithms are proposed that may be explored to improve its accuracy when additional data are obtained. Additional metrics related to eventual sonication parameters should also be identified and explored.