I had my first depression at 19. I received my bipolar diagnosis at 29. Ten years of struggling without fully understanding what was happening.
That is why the question of diagnostic delay feels so personal. When Ishika Agrawal, founder of Nocturnal Neuro, describes an average wait of around 10 years for an accurate bipolar diagnosis, I recognize my own experience. Behind that number are years of uncertainty, treatment decisions, and lives that cannot simply be put on hold.
Could the next big improvement in bipolar care come not only from a new medication, but from better information about our brains and daily lives?
Artificial intelligence, EEG, and other forms of neurotechnology are opening that possibility. But there is a big difference between finding an interesting pattern in data and building something that genuinely helps a patient. We need both hope and evidence.
Psychiatric diagnosis still relies heavily on conversations, clinical observation, and a person’s account of their experiences. The Diagnostic and Statistical Manual of Mental Disorders, usually called the DSM, provides criteria that clinicians use to assess mental health conditions.
Those conversations matter. The problem is how much they are being asked to carry.
Imagine trying to explain your life, your symptoms, and the changes in your mood during an appointment lasting an hour or 90 minutes. You are describing everything from your perspective, using what you remember and what you understand about yourself at that moment. The clinician then has to make sense of that picture.
Ishika points to substantial misdiagnosis between depression and bipolar disorder, citing a figure of about 40%. These estimates are part of her argument for better diagnostic tools, not a prediction of what will happen to any particular person. The underlying concern is clear: too many people spend too long without an accurate explanation for their symptoms.
Nocturnal Neuro focuses on mood disorders, particularly depression and bipolar types 1 and 2. Even within that narrower area, distinguishing conditions is difficult.
Symptoms can overlap across diagnoses. Some people live with more than one mental health condition, which makes the picture more complicated. Researchers also consider relationships across mood and psychotic disorders, rather than assuming every experience belongs in a completely separate box.
That complexity helps explain why a promising diagnostic model cannot simply be rushed into practice. It must work beyond a tidy research dataset, with people whose experiences are messy, overlapping, and changing.
The aim is not to replace the patient’s story with a measurement. It is to give that story more support.
A biomarker is a measurable biological signal that researchers can investigate for a relationship with a condition or its symptoms. Ishika’s work explores whether brain activity could provide useful signals for distinguishing mood disorders.
Her background is in engineering, computer science, and machine learning. Her own mental health experiences helped motivate her to apply those skills to psychiatry. Through Nocturnal Neuro, she is developing a wearable headband and analytical tools intended to support better diagnosis.
The research question is straightforward: can measurements reveal differences that are difficult to identify from conversation alone?
Possible sources of information discussed include:
Machine learning can help researchers search these datasets for patterns. But a pattern is only a starting point. It still needs to be tested to establish whether it is reliable and useful for clinical decisions.
Electroencephalography, or EEG, records electrical activity from the brain using electrodes placed on the scalp. It is already used in neurological care, including the assessment of epilepsy and seizures.
Unlike an MRI image that can show brain structure, EEG captures patterns of electrical activity over time. Those recordings appear as waveforms on a computer screen. Even blinking can produce visible changes in the signal.
Ishika describes two approaches to collecting it:
In the observational study she describes, participants wear a dry EEG headset for about five to ten minutes. The team collects recordings from people with bipolar disorder, people with depression, and people without either condition, then investigates whether meaningful differences exist between the groups.
That distinction is important: this is research into a possible diagnostic aid, not evidence that a short EEG recording can already give someone a definitive bipolar diagnosis. The wearable headband is a development goal, while the study is part of the work needed to assess the underlying idea.
Even an accurate diagnosis does not automatically create a coordinated recovery plan.
A person may have a psychiatrist, a psychologist, an inpatient team, and perhaps a case manager. Yet those professionals can end up working separately. One knows about medication, another knows about recent symptoms, and someone else is trying to help with the practical consequences of taking time away from school or work.
Meanwhile, the patient becomes the person responsible for connecting everything.
I recognize that gap between hospital and home. In hospital, you may have access to several resources. At home, much of the management can fall back on you.
A more connected system would help the care team understand:
Shared information could make care more efficient and reduce some of the workload on professionals. But collecting data is not enough. It has to reach the right people and help them coordinate around the person receiving care.
Why do some therapists embrace mood-tracking apps while diagnostic AI faces much more resistance?
One useful distinction is between mental well-being and treating a mental illness.
Wellness tools can help people reflect on stress, track mood, journal, or pay attention to daily routines. Wearables such as Oura rings and Apple Watches sit within that broader technology landscape.
For many therapists, an app is a practical update to something they already do. Ishika remembers being given an Excel sheet with dates and emotion ratings by her first therapist. Replacing that sheet with an app makes an established process easier.
A tool that influences a diagnosis or treatment decision has a different responsibility. Researchers must show that its results are meaningful, address potential bias, and account for overlapping conditions.
Some caution is necessary. At the same time, the limitations of existing care are a reason to keep investigating, not a reason to stop.
AI tools can seem to change overnight. Clinical research does not move at that pace.
Studies that follow psychiatric symptoms over time may take years to complete. Results then have to be analyzed, published, and translated into something that can be used responsibly. A lack of headlines does not mean nothing is happening.
Ishika describes early-stage companies receiving support from organizations such as the National Institutes of Health and the National Science Foundation to validate ideas and conduct studies.
She also discusses a multi-institution bipolar research effort involving universities including Harvard and UC San Diego. In the study she describes, participants are followed for approximately six months, with researchers gathering MRI data, blood samples, and cognitive assessments.
She is careful not to claim knowledge of that study’s exact hypothesis. Its relevance here is the breadth of information being collected to investigate relationships between biology, bipolar disorder, and symptoms.
Healthcare technology cannot follow the same release cycle as an ordinary software app. A mistake can influence someone’s care.
The discussion touches on the FDA’s three medical-device classes and how regulatory requirements vary with the device and its risks. Ishika describes Nocturnal Neuro’s software as being developed with a Class II pathway in mind, rather than presenting it as an already approved diagnostic product.
The broader challenge is to allow useful innovation while requiring appropriate evidence and safeguards. That applies to software and machine learning as well as physical devices.
Clinical trials can sound frightening. It is easy to imagine an experimental drug and a dramatic science-fiction scenario. Many studies are much less dramatic, with carefully specified procedures and substantial administrative oversight.
Before deciding whether to participate, it helps to understand the difference between observational and interventional research.
An observational study collects information without assigning a treatment intervention to test its effects.
Examples discussed include EEG recordings, MRI scans, blood draws, cognitive tests, and symptom assessments. Nocturnal Neuro’s short EEG recording study is presented as observational because its purpose is to gather and compare data, not to change participants’ treatment.
Observational does not mean that every procedure is free of risk or burden. The specific activities, time commitment, discomforts, and privacy considerations still need to be explained.
An interventional trial tests an intervention, such as a medication or a change in treatment. A ketamine trial is one example raised in the conversation.
That creates a different set of questions about possible benefits, side effects, and implications for current care. The relevant comparison is not simply “research is safe” or “research is dangerous.” It is the risk and potential value of this particular study for this particular person.
I would like research navigation to offer that kind of independent support: helping someone understand their options without pushing them toward one particular trial.
If we want people to share sensitive information, we have to explain what happens to it.
Ishika describes spending months working through study approval, protocols, consent, privacy, and security. In her experience, both the protocol and the consent documentation were substantial, with detailed questions about participants’ rights and data handling.
A meaningful explanation should address:
A vague promise that information might be shared with future collaborators is not the same as a clear account of its intended use.
Research oversight is an important safeguard, but it is not a reason to stop asking questions. Trust comes from transparency, understandable communication, and respecting the limits of consent.
Participants may also receive compensation for their time. Ishika mentions amounts such as $50 or $100, depending on the study’s demands. Others may be motivated by the possibility of helping people with similar experiences in the future. Neither compensation nor participation guarantees a personal treatment benefit.
Getting a diagnosis can be a long journey. Finding a treatment that works can be another one.
Medication decisions often involve trying a treatment, assessing its effects and side effects, and adjusting the plan with a psychiatrist. Repeated attempts can consume months or years.
Online accounts make the uncertainty obvious. Search for the same medication combination and you may find one person describing it as life-changing and another describing a terrible experience. Those are real experiences, but they do not tell you how your own body will respond.
Precision psychiatry aims to make treatment more individual by bringing together clinical history, symptoms, and relevant biological information.
Ishika raises genetic testing as one potential source of information about medication suitability and adverse effects. Lamotrigine, also known by the brand name Lamictal, comes up as an example of why medication safety and gradual dose changes matter.
It is important not to turn that research possibility into a dosing shortcut. A genetic result is not a guarantee that a medication will work or that it can be increased quickly. Medication selection and titration should remain with the prescribing clinician.
The goal is better-informed decisions, not replacing one uncertain rule with another.
I can imagine an appointment where the psychiatrist has more context before we even sit down together:
That is a vision for care, not a description of a universally available service. Its value would be in supporting a more informed conversation and a more personalized recovery plan.
The future of psychiatric innovation is not limited to drug discovery.
At a neural-engineering conference, Ishika encountered companies working on objective symptom tracking and brain-stimulation approaches. The discussion includes deep brain stimulation and technologies involving magnetic stimulation or ultrasound. These are different approaches, not interchangeable names for one treatment.
Some of this work targets difficult problems such as treatment-resistant depression. It raises a reasonable question: could a device help where existing treatments have not?
It also raises a reasonable hesitation. Before considering a technology for ourselves or someone close to us, we want to understand the studies, the risks, and why we should trust the people offering it.
Being hopeful does not require being uncritical.
For all her enthusiasm about machine learning, Ishika is skeptical of the hype around virtual AI therapists. Her position is clear: technology should not replace a qualified human therapist.
That matters because mental health care involves precisely the things that are hardest to reduce to data: feelings, emotions, body language, and the relationship between a person and their clinician.
An algorithm may identify a measurable pattern. It does not follow that it understands what an experience means to the person having it.
There is also a wider issue of who gets to shape care. Governments, pharmaceutical companies, hospitals, and insurers have significant influence. Patients should not be brought in only after the major decisions have already been made.
Better technology should strengthen the patient’s voice, not make it easier to ignore.
Finding a correlation is not the finish line.
Ishika argues that companies need separate studies to establish whether their products improve outcomes when added to the care people already receive.
That is a more demanding question than whether a model performs well on a dataset. It asks whether the technology makes a useful difference in someone’s life.
For founders, that means keeping patient needs, privacy, and security at the center of decisions. For patients, it means we can ask for evidence without giving up hope.
When asked what she would change in mental health with a magic wand, Ishika turns from engineering to acceptance.
She describes the difficulty of accepting that she might need to do things differently: make time for therapy, medication, exercise, or coping strategies such as breathing and meditation.
Many of us recognize that struggle. Shame can make it harder to acknowledge what we need. Acceptance can help us build a routine that supports us, rather than repeatedly fighting the fact that support is necessary.
That does not mean acceptance is a cure or that recovery depends on having the right attitude. It means recognizing your needs without treating them as a personal failure.
You may need to change how you organize your days. That does not mean you have to abandon what you want from your life.
Better data, more coordinated care, and carefully tested tools could help shorten the uncertainty that so many people with bipolar disorder experience. But the purpose remains human: less suffering, clearer decisions, and more room to build a meaningful life.
To explore mental health research opportunities, visit HopeStage’s research page. For more resources on living with bipolarity, visit HopeStage.
