DOCUMENT

6 Headlines The promise and perils of AI in neuroscience Artificial Intelligence (AI) has the potential to transform the diagnosis and treatment of disease. Young neuroscientist Dr Narun Pat says while the technology is a milestone in medicine, we must proceed with caution. I t’s no surprise that AI was the Collins Dictionary word of the year for 2023. In the past 12 months the term AI – a catchall for artificial intelligence, big data and machine learning – has made an evolutionary leap from the desks of computer scientists to everyday use in multiple industries, with neurology no exception. According to Collins, it is considered to be the next great technological revolution. AI is the science of making machines that are programmed to think like humans, but unlike us, AI can also process vast amounts of data. In medical research, being able to quickly analyse massive patient data will undoubtedly lead to major discoveries and advances in patient care. Dr Narun Pat is a cognitive neuroscientist in the Department of Psychology, University of Otago, and an expert on neuroimaging. As an early career neuroscientist, he is excited about AI’s potential to transform healthcare in his lifetime. “Big data and machine learning are creating a paradigm shift in neurology.” He leads a laboratory that uses neuroimaging methods (such as fMRI and EEG) along with modern data science tools (such as big data, machine learning and computational modelling) to better understand the brain and disease. “AI and MRI big data has led to the development of MRI- based biomarkers, found to be useful for disorders such as Alzheimer’s and schizophrenia. MRI-based biomarkers from these technologies can help with risk stratification, earlier detection, diagnoses and prognoses,” Dr Pat says. But before AI accelerates research in his field, he wants to get one step ahead of it to ensure it benefits all patients. This is because bias in vast data sets is emerging as a real concern. “Machine learning, even when built from seemingly large data sets, can suffer from an ethnic bias if the training data is not diverse enough. “Brain-MRI big data have largely been collected from participants of European descent, setting the scene for ethnic biases,” Dr Pat explains. “This means that the MRI-based biomarkers, built from participants of European descent, are more likely to perform better on participants of European descent than those of other ethnicities.” AI is perfectly placed to rapidly advance our knowledge of biomarkers by comparing vast numbers of MRI-images. Biomarkers are an incredibly important tool for doctors to detect and confirm disease. They can range from your pulse or temperature, to a nuance on an MRI brain image that indicates a tumour. Dr Pat is in the process of number crunching MRI images from 70,000 individuals from around the world. Once the ethnic bias is investigated and perhaps accounted for, the data set presents an enormous opportunity to find new and more precise biomarkers for brain disorders. The project to investigate and correct for ethnic bias is likely to take around three years, even using AI technology. It received a large project grant in the Neurological Foundation’s latest grant round to see it through. Dr Pat has secured almost a petabyte of space to house the database (that’s one billion megabytes, equivalent to the storage capacity of 7,812,500 new iPhones). The number crunching will be handled by New Zealand’s most powerful supercomputer, NeSI (New Zealand eScience Infrastructure). His research team includes experts from various fields: cognitive neuroscience (himself and his student, Alina Tetereva), medical physics (Associate Professor Tracy Melzer from New Zealand Brain Research Institute in Christchurch), data science (Associate Professor Jeremiah Deng from the School of Computing, University of Otago) and genetics (Dr Ric Anney from Cardiff University in the UK).