Apple M3 Pro

Apple Vision Pro利用虚拟现实和增强现实技术，能够为医学教育、临床诊断与治疗以及医疗管理等领域带来革命性的变革和贡献，推动元宇宙医学的发展和应用。然而，实际应用Apple Vision Pro的过程中也需要解决技术和伦理挑战，包括数据隐私、安全性问题、医疗责任和法律法规考虑等。本文旨在探索Apple Vision Pro在元宇宙医学场景中的应用，以期推广Apple Vision Pro在医学领域的应用，促进医学教育、临床实践和医疗管理的全面进步。

Jason Kim, Jalen Chuang, Daniel Genkin, Y. Yarom

Tao Lan, Sichun Liu, Yihe Dai, Jia Luo, Jiang Han, Yun Jin

Laparoscopic ultrasound-guided liver microwave ablation requires precise navigation and spatial accuracy. We developed an image fusion navigation system that integrates laparoscopic, ultrasound, and 3D liver model images into a unified real-time visualization. The Apple Vision Pro mixed reality device projects all essential image information into the surgeon’s field of view in real-time. This system reduces cognitive load and enhances surgical precision and efficiency. Comparative experiments showed a significant improvement in puncture accuracy under AVP guidance (success rate of 90%) compared to traditional methods (42.5%), benefiting both novice and experienced surgeons. According to the NASA Task Load Index evaluation, the system also reduced the workload of surgeons. In eight patients, ablation was successful with minimal blood loss, no major complications, and rapid recovery. Despite challenges such as cost and fatigue, these results highlight the potential of mixed reality technology to improve spatial navigation, reduce cognitive demands, and optimize complex surgical procedures.

Mehiar Ahmed Tarek Fikry, M. Nagy, S. Fahmy

Aim : The aim of the study was to compare the changes in angle of curvature, transportation and centering ratio after instrumentation in M3 Pro Gold system and M Pro system Pro Taper Universal system was used as gold standard and reference for comparison. Materials and Methods : Thirty human mandibular first molars were selected. Teeth were randomly allocated into three groups (n= 10): Group 1: M3 Pro Gold; Group 2: M Pro; Group 3: Pro Taper Universal. Cone Beam Computed Tomography (CBCT) were taken before and after preparation measurements were made using On Demand software. Evaluation of angle of curvature was done using Schneider’s method. Transportation and centering ratio were evaluated using Gambill’s method. Comparison between 3 groups was performed by using One Way ANOVA test followed by Tukey`s Post hoc test for multiple comparisons. Results : After root canal preparation significant difference in the change of canal curvature values were found, Pro Taper Universal showed the least change and M Pro showed the highest change (p = 0.0001), while canal transportation results showed significant difference between all groups at all levels as M3 Pro Gold showed the least transportation and M Pro showed the highest transportation (P = 0.0001). Centering ratio results showed significant difference between all groups at all levels as M3 Pro Gold was better centralized in root canal in comparison to M Pro and Pro Taper Universal (p = 0.0001). Conclusion : M3 Pro Gold better respected the original canal anatomy and better conserved root dentine structure as compared to M Pro and Pro Taper Universal.

Prabod Rathnayaka, Fabian Waschkowski, Lukas Wesemann

We present BaseRT, a native Metal inference runtime for large language models (LLMs) on Apple Silicon, and report the highest inference throughput on this hardware to date. Existing runtimes, including llama.cpp and MLX-based frameworks, incur overhead from abstractions not designed for Metal's execution model or Apple Silicon's unified memory topology. By building natively on Metal with chip-specific kernel fusion, unified memory-aware optimisation, and custom dispatch logic, BaseRT recovers performance that framework-based approaches leave on the table. BaseRT supports a wide range of model families across eight quantisation formats (Q2 to FP16) on all Apple M-series devices. In this paper, we evaluate the Qwen3, Llama 3.2, and Gemma 4 families at Q4 and Q8 quantisation on M3 and M4 Pro devices. BaseRT achieves up to 1.56x higher decode throughput than llama.cpp and up to 1.35x higher than MLX, with substantially larger margins on prefill for mixture-of-experts models, delivering consistent best-in-class throughput from sub-1B to 30B parameter models. These results establish Apple Silicon as a more capable inference platform than previously reported, with direct implications for the emerging edge inference paradigm: as privacy requirements, latency constraints, and cloud cost pressures drive inference toward on-device deployment, performance-optimised local runtimes are a critical enabling layer for this transition. BaseRT is publicly available at https://github.com/basecompute/baseRT