Robot-assisted stereotactic drainage integrating AI-aided segmentation versus conventional craniotomy for severe primary brainstem hemorrhage: a retrospective exploratory cohort study

Primary brainstem hemorrhage (PBSH) is characterized by acute onset, rapid progression, and exceptionally high mortality. Despite advances in microneurosurgery and refined anatomical safe-entry zone mapping, the optimal management of PBSH remains controversial. This study aimed to evaluate the feasibility and observational outcomes of AI-assisted, robot-guided stereotactic drainage compared with conventional craniotomy in patients with severe PBSH. This single-center retrospective cohort study enrolled 34 adult patients with severe PBSH (hematoma volume ≥ 5 mL, GCS ≤ 8) treated between May 2024 and May 2025. Thirteen patients underwent AI-assisted robot-guided stereotactic drainage, and 21 received conventional craniotomy evacuation. The primary outcome was 90-day mortality (GOS = 1). Secondary outcomes included operative time, 30-day mortality, and drainage duration. Baseline characteristics were statistically comparable between the two groups. Operative time was significantly shorter in the robotic drainage group (2.37 ± 1.01 h vs 6.90 ± 2.02 h, p < 0.001). The 30-day mortality was 15.4% (2/13) in the robotic group versus 33.3% (7/21) in the craniotomy group (p = 0.27). At 90 days, mortality was 23.1% (3/13) versus 66.7% (14/21) (p = 0.035). Kaplan-Meier survival analysis demonstrated a statistically significant survival advantage for the robotic approach in the unadjusted analysis (log-rank p = 0.0180). However, no patient in the robotic group achieved functional independence (GOS 4–5) within 90 days; most survivors remained in a vegetative state or severely disabled (GOS 2–3). Robot-assisted stereotactic drainage with AI-assisted initial segmentation was associated with shorter operative time and lower 90-day mortality than craniotomy in severe PBSH; no procedure-related rebleeding occurred, although adverse events were not systematically monitored and quantitative targeting accuracy could not be assessed. These exploratory findings support technical feasibility and a potential short-term survival benefit but require prospective multicenter validation with longer follow-up, quantitative endpoints, and patient-centered outcomes.