NCCN adds Modeyso (dordaviprone) for H3 K27M diffuse midline glioma

Now a Category 2A recommendation, marking the first systemic therapy in guidelines for this tumor.

FDA approves Modeyso (dordaviprone) for H3 K27M-mutant glioma

Accelerated approval for adults and children with progressing H3 K27M-mutant diffuse midline gliomas.

FDA approves Leqembi Iqlik autoinjector for Alzheimer’s

Subcutaneous 60 mg/5 mL device enables at-home lecanemab administration.

FDA grants accelerated approval to FORZINITY (elamipretide)

First treatment for Barth syndrome, a rare pediatric mitochondrial and neuromuscular disorder.

EMA CHMP positive opinion for Aqneursa (N-acetylleucine)

Backed for neurological symptoms of Niemann–Pick type C in adults and children ≥6 years.

FDA priority review for apitegromab in SMA

Myostatin inhibitor under review, decision due September 2025.

UC Davis launches stem-cell gene therapy for Angelman syndrome

Program engineers bone marrow stem cells to deliver UBE3A protein to the brain.

Moberg Analytics launches Moberg Clinical Platform

AI integrates EEG, imaging, and clinical data for brain injury management.

Newronika advances AlphaDBS adaptive DBS system

Keith Carlton appointed Chairman as U.S. pivotal trial preparation continues.

Brainomix e-Stroke AI triples stroke recovery rates in UK rollout

Patients treated with AI triage had threefold higher recovery rates vs standard care.

BrainScope expands EEG platform

Provides biomarkers for TBI, stroke triage, and early Alzheimer’s detection.

EpiSMART continual learning seizure detector (Shahbazinia et al.)

Continual-learning algorithm adapts daily to patient EEG, boosting seizure detection accuracy with minimal labeled data.

EEG-based BCIs review (Barbera et al.)

The September 2025 arXiv review analyzed >200 recent studies and showed how AI is boosting EEG-based brain–computer interfaces. Transformer and convolutional models reduced classification error rates by ~20% compared with older CNN pipelines, and transfer learning allowed cross-patient generalization with up to 70% accuracy in noisy data. Key pillars: advanced preprocessing, unsupervised feature embeddings, and continual adaptation.

In practical applications, AI-powered EEG BCIs are achieving 10–15% word error rates for imagined speech decoding (vs 30–40% with legacy methods). Motor control tasks using RNNs and transformers improved cursor stability by 25–30% in non-invasive EEG, narrowing the gap with invasive ECoG systems. Communication BCIs now routinely surpass 90% accuracy in binary yes/no tasks, highlighting tangible clinical utility for patients with severe paralysis.