Scribd
Upload
8 views5 pages

Problem Statement

TeamX has developed ClearMed, an AI-powered medical summarization tool designed to create two types of summaries: one for clinicians with detailed medical information and another for patients in simple language. The system ensures that every summary line is linked to the original report for verification and includes safety disclaimers for sensitive information. ClearMed operates locally without the need for cloud services, making it a unique and efficient solution for summarizing complex medical documents.

Uploaded by

rishabh09454
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views5 pages

Problem Statement

TeamX has developed ClearMed, an AI-powered medical summarization tool designed to create two types of summaries: one for clinicians with detailed medical information and another for patients in simple language. The system ensures that every summary line is linked to the original report for verification and includes safety disclaimers for sensitive information. ClearMed operates locally without the need for cloud services, making it a unique and efficient solution for summarizing complex medical documents.

Uploaded by

rishabh09454
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Team Name: TeamX

Challenge Accepted: Medical Summarization Challenge


Project Title: ClearMed
Team Details: Rishabh Gupta (Team Leader)
Sachin Gupta
Rishabh Dubey
Ranjan Yadav

From College: ABES Engineering College, Ghaziabad


Problem statement:
Medical documents are often long and complex.
Doctors need clinically relevant summaries, while patients need easy-to-understand summaries.
Existing tools rely on LLMs with APIs, but here we must build a custom lightweight summarization model.

Solution:
ClearMed (AI System) is an AI-powered solution that can read long medical reports or Q&A documents and generate
two types of summaries:

Clinician Focused Mode → Gives a summary with correct medical terms and detailed clinical information.

Patient Friendly Mode → Gives a summary in simple, clear language without medical jargon.

The AI makes sure:

Every line in the summary is linked to the original report, so users can verify the source.

It checks for risky or sensitive statements (like dosages or absolute instructions) and adds a safety disclaimer.

In Patient Mode, it automatically translates complex medical terms into simple words while keeping the meaning correct.
Why Our AI is Unique

Simple language converter: Two modes (Doctor & Patient):


Complex medical terms are The AI has two separate
automatically replaced outputs – one keeps full
with easy-to-understand medical details for doctors,
words for patients. the other simplifies terms for
patients.

Runs locally: The model is Trustworthy summaries:


small and fast, so it works Every line in the summary
on a normal laptop without is linked back to the original
cloud or external APIs. report so nothing is made
up.

Safe to use: A built-in checker scans for risky advice (like


dosages) and adds warnings/disclaimers
System Architecture
Ingest (Input Handling)
If the PDF is already searchable → parse text directly.
If the PDF is scanned → run OCR to extract text and preserve layout.

Sectionizer
Automatically detects key sections: History, Symptoms, Diagnosis,
Treatment, Follow-up.

Medical NER + Ontology Linker


Identifies medical terms and links them to standard IDs (UMLS).
Builds a glossary for simpler explanation later.

Summarizer (Dual-Head Model)


Uses one shared encoder with two decoders:
Clinician view (Decoder-C): detailed, technical summary.
Patient view (Decoder-P): easy-to-read summary.

Web UI
Side-by-side clinician & patient summaries.
Evidence highlighting, disclaimers, and PDF export option.
Technical blueprint
Pre-processing
PDF & OCR: PyMuPDF / pdfminer, Tesseract
Text Cleaning & NLP: spaCy
Section & Entity Detection
Models: Lightweight Transformer / BiLSTM
NER + Ontology: Transformer-based NER, UMLS
mapping
Evidence Retrieval
Search: BM25 (sparse) + Dense embeddings
(MiniLM)

Summarization Model
Framework: PyTorch
Architecture: Distilled Transformer encoder + Dual
Post-processing decoders (Clinician & Patient)
Glossary-based jargon simplifier
Risk classifier for unsafe claims

Application Layer
Frontend: Streamlit or Flask (with JS for
Optimization highlighting)
Export: PDF reports
Model distillation & quantization for
efficiency

You might also like