Listen to this story
Revolutions happen when humans and machines work in tandem. But why is there a need for neurologists in designing AI algorithms, when we have engineers for it? The answer is to be free of bias, which cannot happen without neuroscientists stepping in.
Scenario without neuro experts
AI is moving brainwards and there has been consistent growth in the number of respondents implementing AI – which is nearly 90%. Studies have revealed that an average investment of $40 million has been recorded so far for the next five years’ AI plan. The need for the involvement of neurologists arose from previous AI algorithms that were biased in terms of gender, race, and other variables. This bias has raised questions about engineers designing the AI models for deployment in clinical trials and the overall healthcare settings.
A report from Chilmark, a global research and advisory firm providing healthcare IT solutions, pointed towards the discriminatory first-year COVID-19 pandemic algorithms, which nullified the accuracy of X-rays and CT scans. The most common bias comes from the inaccuracy of algorithms not being able to define the target population purely designed by engineers.
AI algorithm designed under expert supervision
The focus should be on designing it in a way to constantly assist doctors in updating information from old or recent published journals, articles, or textbooks to improve patient care. This simply means extracting important information from extensive patient data to calculate the probability of disease outcomes. And neuro experts can pretty much understand the nuances of designing genetic data and structural imaging, thus allowing more clarity on the exact patient problem. Methods such as NLP further help extract the unstructured (medical journals) data and turn it into machine-readable data, which ML can later scrutinise, if trained properly under clinical supervision, not just engineering design. Otherwise, it can have contradictory effects on the complete data analysis.
Neuro experts safeguarding the future of AI imaging
The process of filtering out images based on deep learning is now widely acknowledged by various scientists and experts. This algorithm helps in analysing strokes and brain haemorrhages through images at an early stage. If the computer detects any problem in their image analysis, their case will immediately go to the priority segment. And vice versa on the priority list, based on the algorithm detection of image analysis. However, if not designed well, the AI control over image quality, report structure, and computer-aided classification would suffer.
Such deep learning algorithms can enhance MRI image quality, which neuro experts can properly design, and thus ensure improved neuroimaging technology, better preoperative evaluation, and network analysis. Glioma grading by resting-state fMRI, presurgical localisation of the eloquent cortex, and epileptic focus are a few examples of conditioned AI imaging. In addition, experts designing AI can help sort out the piled-up data sets produced by modern neuroscience tools, such as real-time imaging and multi-electrode arrays. Along with this, a correct AI algorithm can help incorporate more knowledge discovery, data mining, segmentation, pattern recognition, graphic visualisation, and other crucial fields yet to be invented with time.
Nishit Agarwal, a biomedical data science engineer from USA, spoke to Analytics India Magazine about the need to bring clinical neuroscience experts to design algorithms, thus not giving full reins to engineers. He is currently working for Medidata, and previously worked as a computational neuroscientist with expertise in creating smart algorithms for processing physiological data collected from the human body using various wearable sensors.
AIM: What do you think accelerates AI penetration in healthcare?
Agarwal: The past decade has been quite unpredictable for the world of artificial intelligence and medicine. With more and more medical practitioners getting familiar with machine learning in healthcare, countless innovations in symptom tracking, disease diagnosis, and precision medicine are being developed. Post pandemic, medical-grade wearable devices have paved the way for home clinical trials. At last, AI and machine learning techniques have enabled neuroscience research to capture the underlying pattern in neural data, accelerating its significance all over. Today, startups like Neuralink, Neurable, Next Mind, Emotiv and many others are investing millions in developing AI algorithms.
AIM: Why do we need clinical neuroscience experts in AI design rather than engineers?
Agarwal: The data lacks in quality without the presence of clinical experts designing it. Algorithms depend on factual data, and the quality of this data depends directly on the experiments conducted and the quality of the electrodes. Unfortunately, so far, there has been minimal input from experts in designing these AI algorithms, showcasing a huge gap in domain knowledge. Neuroscientists can greatly affect the quality of information collected from the data by helping design holistic experiments. Their expertise eliminates the obstacles and biases for engineers in achieving the goal of creating algorithms. Extensive evaluation and constant supervision from clinical neuroscience experts could help data scientists like me change the way we interact with the world directly from our brains.
AIM: What is the future of this problem?
Agarwal: The use of AI in neuroscience is still at an early stage, but the NeuroAI community is growing rapidly. More and more engineers, data scientists, and AI enthusiasts are coming together to solve the challenge of understanding the brain and using AI for groundbreaking innovations in neuroscience. Let’s hope to change the perspective of developing it and give control to clinical experts in shaping the AI algorithms for a bias-free world and usage.
AIM: Do you think neurologists are shaping AI?
Agarwal: Neurologists are shaping AI to give better clarity to the human brain by mimicking its mental functions. Furthermore, they focus on uncovering a massive amount of relevant information through AI.