RT @SBeniczky: Take on the 2025 Seizure Detection Challenge and contribute to the future of epilepsy research!.Submission deadline: Februar….

Presenting this at @mlforhc in Toronto this morning! Come and see poster 182 :). Paper:

A huge thank you to @DorienBlooijs, @dora_hermes and the rest of the team for making the dataset openly available ( as well as to every patient involved. Project code can be found here:

There’s still a long way to go before these models can be clinically useful. Future work will include:.- Exploring pre-training techniques.- Focusing on outcome rather than the SOZ alone.- Model explainability .- Generalising across centres/recording modalities.…and more!. n/n.

We also found that the Transformers didn’t just benefit from seeing more channels during inference, requiring only 4 channels to match the performance of the CNN (which used 37). This demonstrated the effectiveness of cross-channel attention. 5/n

Results:.- In this cohort of patients that underwent 0.2 Hz SPES, the divergent paradigm performed poorly. - The convergent paradigm led to better results, as the literature suggested. - Switching to a Transformer led to a statistically significant improvement. 4/n

2) Generalising to unseen patients: most techniques assume that the input is a fixed number of consistently ordered channels, which isn’t the case in iEEG. For this, we implemented encoder-only Transformers to extract features across arbitrary electrode numbers/placements. 3/n

1) Analysis paradigm: shall we classify a site from the responses that stimulating it evokes elsewhere (divergent), or from the responses at this site upon stimulation elsewhere (convergent)?. We implemented an existing CNN approach and compared both paradigms. 2/n

Excited to share this preprint along with @roschkoenig, @aswinchari and the rest of the team!. This work focuses on two considerations when localising the Seizure Onset Zone (SOZ) from responses to Single-Pulse Electrical Stimulation (SPES). 1/n.

RT @EllershawSimon: 1/n Doctors spend too long doing paperwork. We set out to reduce this using (shock) LLMs prompted by clinical guideline….

RT @aswinchari: Our #EpilepsySurgery research group @GOSH_NeuroSurg & @DNP_ICH has been alive for many years but so far, nameless and facel….

RT @roschkoenig: So pleased I have the chance to get involved with this - excited about thinking through national strategies to maximise th….

RT @yusuf_i_k: The other day I made some charts 📈 for the @polycrisis project. They were for a great piece on the globalisation debate, wri….

RT @peternealtaylor: Normative brain mapping using scalp EEG and potential clinical application ..

RT @peternealtaylor: Interested in learning about applying computational approaches to questions in clinical neurology? We are organising a….

RT @EpilepsiaJourn: Stimulation to probe, excite, and inhibit the epileptic brain. #electricalstimulation #diagnosi….

RT @EpilepsiaJourn: Passive and active markers of cortical excitability in epilepsy . #epilepsy #EEG #biomarkers #H….

RT @IWGBUoL: 🚨 STOP FIRE AND REHIRE REDUNDANCIES AT UCL 🚨. @UCL and @BidvestNoonan have told outsourced security guards that they will all….

RT @AdritoDas: Registration for the PitVis EndoVis 2023 MICCAI Challenge is open! The challenge pertains to workflow recognition in endosco….

RT @roryjpiper: 📢 Interested in doing a PhD in clinical neuroimaging at @DNP_ICH / @GOSH_NeuroSurg ?! THE EPSRC i4health PhD programme is o….