A quiet revolution is reshaping the diagnosis of mental illnesses. Diagnostic practices are shifting: Patients no longer must visit hospitals for tests; instead, they can take mental health screenings at home as easily as measuring blood pressure. Medical professionals are turning these “invisible ailments” into measurable signals by letting the brain communicate its inner state.
When viewing the same oil painting, healthy people naturally fix their gaze on core subjects such as human faces. Patients with depression, by contrast, let their attention drift across the entire canvas, failing to distinguish central figures from marginal backgrounds. A research team led by Professor Lü Baoliang and Associate Professor Zheng Weilong from Shanghai Jiao Tong University converts these behavioral disparities into attention heatmaps, visually rendering invisible emotional states.
“Oil paintings feature extremely rich emotional tones that trigger diverse moods in viewers,” Zheng Weilong explained. “They contain varied figures, landscapes and objects. While observing paintings, viewers generate abundant subconscious behavioral traces that our equipment can record.” Eye movements reveal a wealth of subconscious responses. The team combines eye-tracking records with electroencephalogram (EEG) signals, using both physiological and behavioral markers to evaluate a person’s mental status. The team has collected high-quality clinical scale and EEG data from over 6,000 patients diagnosed with depression, alongside datasets from more than 1,300 healthy control participants.
Researchers identified a pattern they name negative bias: when exposed to oil paintings that easily trigger negative emotions, depressed individuals fixate excessively on gloomy visual elements. After training artificial intelligence (AI) models on targeted painting stimuli, the system can distinguish depressed subjects from healthy ones with an accuracy rate above 90%.
These technological advances address a severe global public health crisis. Roughly one billion people worldwide live with various mental disorders, yet more than 75% lack access to effective treatment. Mental illnesses lack objective biological biomarkers. Clinicians still rely primarily on patient interviews and self-reported questionnaires for diagnosis, leading to high subjectivity and frequent misdiagnoses.
Back in 2015, Thomas Insel, former director of the U.S. National Institute of Mental Health, published an appeal in Nature. He argued that mental disorder diagnosis must shift from qualitative assessment to quantitative measurement through a standardized system of objective biological biomarkers. A decade later, the fusion of AI and brain science delivers viable solutions to this longstanding challenge. On May 21st, at the 2026 China Computer Federation Young Elite Forum (YEF2026) held in Mianyang, Sichuan Province, numerous scientists shared their latest research breakthroughs.
The Shanghai Jiao Tong University team led by Zheng Weilong focuses on developing large-scale EEG foundation models. Traditional EEG collection carries exorbitant costs, and high-quality clinically labeled datasets remain scarce. These limitations left earlier models unable to generalize to new diagnostic tasks.
The rise of large language models inspired a new strategy: self-supervised pre-training. Researchers first pre-train models on massive unlabeled EEG datasets to build a foundational knowledge base. The system then conducts joint cross-dataset and cross-task learning before adapting to specific diagnostic scenarios.
The team created the MindCross model, which moves beyond closed-set classification to support open interactive analysis. After inputting an EEG waveform, the model generates visual waveform outputs and delivers clinical diagnostic judgments comparable to professional psychiatrists. Dubbed an “emotional X-ray machine”, the device conducts a 5-10 minute scan that integrates multi-modal physiological signals including EEG, eye movement tracking and facial expression capture. It creates a full “scan” of brain activity and automatically generates detailed reports mapping emotional distribution. Zheng Weilong stated the device’s core mission: enabling early screening, early diagnosis and early intervention for mental illnesses. It provides scientific reminders for people who do not recognize their own psychological distress. The equipment has completed clinical verification in multiple hospitals, and engineers are working to miniaturize it for portable daily use.
Practical, low-cost yet highly accurate diagnostic hardware represents the long-term research goal of Professor Hu Bin’s team at Beijing Institute of Technology. Most conventional EEG systems require full-head electrode arrays, demand lengthy testing sessions and suffer signal interference from hair.
Hu Bin’s clinical trials prove that data collected from just three prefrontal electrodes over several minutes can rapidly differentiate depressive patients from healthy individuals. The technology has undergone clinical testing on approximately 9,000 volunteers across Beijing, Shanghai and other cities. The relevant device has obtained Class II medical device registration certification and is deployed in freshman physical examinations at several universities. The team also built a mobile application that identifies emotional distress via smartphone-based voice and facial expression analysis, reaching an 80% accuracy rate under controlled laboratory settings. Hu Bin acknowledged that background noise, fluctuating lighting and other real-world environmental factors reduce recognition performance in public spaces. “Analysis of eye movements and pupil dilation effectively reflects emotional shifts,” Hu Bin noted. Pupillary responses differ drastically between depressed and healthy people when facing engaging stimuli or experiencing distinct moods.
From this observation, the team put forward the concept of emotional bandwidth, a quantitative metric to measure the range and intensity of emotional fluctuations.
Overseas commercial eye-tracking glasses previously sold for up to 350,000 RMB per unit with locked proprietary interfaces. The team independently developed domestic eye-tracking eyewear, cutting hardware costs down to several thousand RMB while maintaining roughly 80% recognition accuracy. Professor Yang Yun from the School of Artificial Intelligence, Yunnan University gained a pivotal insight during discussions with clinical physicians. Many medical conditions do not rely entirely on medical intervention to heal. Medical treatments only support the human immune system to overcome illness itself. This made him realize that psychological pathogenic factors are widely overlooked in clinical practice. “Some illnesses stem not from physical abnormalities but psychological strain,” Yang learned from collaborating doctors. Untreated psychological disorders produce physical symptoms identical to organic diseases – and often trigger worse physical discomfort.
For this reason, Yang Yun’s research concentrates on depression diagnosis. His team designs stimulus tasks that elicit authentic emotional reactions, while simultaneously capturing multi-dimensional physiological data including facial expressions, electromyography (EMG), EEG, vocal signals and body temperature. AI algorithms synthesize all signals to deliver comprehensive assessments. “Our laboratory clinical trials achieve an accuracy rate exceeding 90%, marking substantial performance improvements,” Yang said. The laboratory has developed four proprietary products. Two consumer-grade versions are already commercially available, and one clinical-grade device is undergoing medical device licensing approval. “Fields with uncharted mechanisms – such as the inner workings of the human mind – stand to gain the greatest transformative value from artificial intelligence,” Yang Yun commented.
The research group led by Lü Baoliang and Zheng Weilong curates SEED, the world’s largest public emotional EEG dataset. Statistics as of March 2026 show the dataset has received more than 10,000 access requests from 3,400+ universities and research institutions spanning 106 countries, supporting countless global mental health studies.
Hu Bin’s research team open-sourced a multi-modal depression dataset covering EEG, eye tracking, voice recordings and other biological signals, free of charge to all global research organizations. Over 1,000 international institutions have downloaded and utilized this dataset to advance worldwide mental health research.
In March 2026, the National Medical Products Administration (NMPA) approved the world’s first invasive brain-computer interface (BCI) medical device for commercial launch. This milestone marks the formal transition of BCI technology from laboratory research to routine clinical application. The breakthrough inspires researchers to pursue a broader objective: reforming mental healthcare. Diagnosis will shift from experience-based subjective judgment to data-driven objective biomarker identification. This quiet technological revolution continues to accelerate worldwide.
Yet quantifying mental illness through digital technology faces steep obstacles, and progress will not be seamless. “Similar to other large foundation models, EEG-BCI large models generate hallucinatory false outputs, and effective mitigation strategies plus refined pre-training tasks remain to be explored,” Zheng Weilong pointed out frankly.
Major challenges persist in integrating objective digital assessment tools with digital therapeutics to build closed-loop intervention systems for mental health management.