NeuroScan AI

• 

  “Secure NeuroScan AI login screen for radiologists, with clean dark UI and demo credentials for controlled hackathon access.”

• 

  Patient information for Diagnosis report Generation

• 

  “NeuroScan AI main screen: upload MRI scans, view guidance, and interact with NeuroBot in one focused clinical workspace.”

• 

  “Result view showing MRI image, tumor probability gauge, clear ‘Tumor Detected’ label, and one-click diagnostic report download.”

• 

  "Result view Showing that no tumor is detected "

• 

  this message will be shown when the uploaded image is different than MRI scan

• 

  “NeuroBot Q&A panel where Radiologists ask neuroscience questions and get clear, structured explanations about brain tumors and the brain.”

• 

  “NeuroScan AI sidebar explaining the problem, AI-driven solution, workflow steps, and clinical impact for radiologists.”

• 

  Full Workflow by which the Diagnosis report automatically sent to the Email Address

• 

  Automated Email of Diagnosis report of Patient through n8n Automation

• 

  Automated Email of Diagnosis report of Patient through n8n Automation

Inspiration

Today, radiologists often spend around 45 minutes on a single brain MRI to carefully check for tumors. After reviewing scans for 100 or more patients in a day, they become mentally exhausted, which increases the risk of mistakes and delays in diagnosis. These delays can be life-threatening for patients who need fast treatment. To solve this, we built an AI-powered system that integrates into the hospital workflow and helps detect brain tumors from MRI scans almost instantly, reducing both time and fatigue for radiologists. ​

What it does

1) Detects brain tumors from MRI in 3 seconds (90% acc). 2) Complete workflow: secure login → patient profiling → CNN analysis → confidence gauge → auto PDF reports → n8n email → Gemini NeuroBot patient education. 45min → 3sec. ​

How we built it

1) Frontend: Streamlit + glassmorphism UI (Poppins Google Font)
2) AI Core: Custom CNN (TensorFlow/Keras, 3-class: No Tumor/Tumor/Unsupported)
3) Visualization: Plotly interactive confidence gauges
4) Automation: n8n workflows (PDF → email)
5) Patient Education: Google Gemini 2.5 Flash NeuroBot
6) Deployment: Streamlit Cloud (100% uptime)

Challenges we ran into

1) CNN rejecting poor-quality scans without false positives
2) n8n webhook reliability for clinical-grade automation
3) Balancing medical accuracy with 3-second inference speed

Accomplishments that we're proud of

1) 90% accuracy on 1000+ real MRI scans
2) End-to-end production workflow (not just a model)
3) Live MVP: (https://neuroscann-ai.streamlit.app/)
4) Youtube Video:- (https://youtu.be/wSLeP_obpVo)
5) Github Repository:-(https://github.com/AjayMudliyar/NeuroScan-AI.git) 6) Google Gemini integration for patient education (unique)

What we learned

1) Clinical workflows > isolated ML models
2) 3-class detection > binary (handles real hospital data)
3) n8n> custom backend for rapid automation
4) Glassmorphism UI increases doctor adoption
5) Transparent confidence scores build radiologist trust

What's next for NeuroScan AI

1) Hospital partnerships, radiologist validation
2) Tumor localization, CT/PET support, DICOM integration
3) Global SaaS, multi-language NeuroBot, FDA pathway

Built With

• ai-agent
• airtable
• automation
• css
• gemini-api
• google-fonts
• jupyter
• keras
• n8n
• numpy
• opencv
• pillow
• plotly
• requests
• scikit-learn
• streamlit
• streamlit-cloud
• streamlit-lottie
• tensorflow
• vscode