Piramidal’s Base Model for Brainwaves Might Significantly Enhance EEG Performance

Artificial Intelligence is being leveraged to process and analyze vast arrays of data, yet its performance varies. This variability is evident within the medical realm, where a new startup named Piramidal is introducing a groundbreaking model designed specifically for the analysis of brain scans.

Co-founders Dimitris Sakellariou and Kris Pahuja have taken note that, despite the widespread use of electroencephalography (EEG) across hospitals, the technology remains divided across various devices and necessitates expert interpretation. They propose a software solution capable of identifying alarming patterns in brain activity consistently, which could significantly enhance the care for patients with neurological disorders, while also alleviating the burden on the medical staff.

Pahuja highlights a critical challenge in neurology intensive care units (ICUs), where nurses monitor EEGs for critical signs but may miss vital cues due to other duties. Incorrect or missed interpretations can mean overlooking serious events like epileptic seizures or strokes, where the expertise of specialized doctors is imperative yet not always immediately available.

After years of exploring the potential of computational analysis in neurology, the founders embarked on creating Piramidal. They’ve identified a pathway to automate EEG data analysis in a manner beneficial to patient care, yet implementation of such technology at the point of need remains complex.

Sakellariou reflects on his direct experience with neurologists to grasp the importance of brainwaves and the technical challenges in creating systems that accurately identify them. Each EEG application prompts the need for a tailored system, requiring fresh data and manual data annotation—a cumbersome process.

This complexity persists, compounded by the variability in EEG systems, hospital IT infrastructures, and data formats, including differences in electrode numbers and placements.

Believing in their methodology, Sakellariou and Pahuja assert their unshared foundational model for EEG interpretation could revolutionize immediate, effective identification of critical brain patterns, bypassing the lengthy process of traditional studies.

The goal isn’t to create an all-encompassing medical tool but, similar to Meta’s Llama, lay the groundwork for advanced capabilities—in this case, understanding brain activity from EEGs. This foundational understanding could then be applied universally across different machines, electrode setups, and patient profiles.

Although unveiling a universally applicable EEG analysis model remains a work in progress, the duo is keen to emphasize that their foundation model is under active development, with aims at large-scale applicability.

Pahuja announced plans for early hospital deployments next year, indicating forthcoming pilots within ICUs to validate the model’s applicability in real-world settings, signifying a major step towards clinical integration.

While the foundation model will require specific adjustments for particular uses, Piramidal intends to handle initial customizations in-house, diverging from the common AI industry practice of profiting from API usage.

Sakellariou conveyed confidence in the preeminent capabilities of their model over any newly developed variant, highlighting the unprecedented scale of their EEG model.

Progress hinges on securing two critical resources: investment and data. The startup has initiated their journey with a $6 million seed financing round. They’ve also begun compiling a vast dataset, utilizing both open source and collaborative hospital data to refine their model further.

Such collaborations with hospitals promise a wealth of EEG data, propelling the model’s capabilities significantly ahead, potentially surpassing human analytical performance.

While surpassing human capabilities is a future ambition, the immediate focus remains on enhancing the quality of neurological care through the upcoming ICU pilots, which will rigorously test and document the technology’s effectiveness and reliability.