Model Testing

At c-riht, we evaluate each AI model's latest release pertaining to the medical and healthcare context.

1. The testing will be done at phases using different methodologies under different categories.

2. We test models locally that means within our system. For bigger models, we use cloud testing under secure environments.

3. We encourage participants to participate the live demo of Model's evaluation which will be posted in the Events section https://www.c-riht.org/innovdoc

4. We won't disturb in emails For Updates join the channel:  https://whatsapp.com/channel/0029VavOyZW7oQhgVdnzqY1k

5. Following are the generalized testing categories that we use for public demonstration. However our testing and evaluation involves deep technological metrics.

For one vs one data sharing, data testing, digital health product building or any help us reach you here

Testing Categories

Below are the functional areas we will probe during the live session. Each block describes a concrete sub‑task participants can submit as a prompt.

Clinical Knowledge

• Disease diagnosis & treatment

• Medical terminology & calculations

• Anatomy & pathophysiology

Clinical Reasoning

• Differential‑diagnosis generation

• Risk stratification & triage

• Evidence‑based decision making

Drug Safety & Interactions

• Polypharmacy management

• Drug‑drug interaction detection

• Dosing calculations & adjustments

Radiology (2‑D) Classification

• Chest‑X‑ray multi‑label prediction

• CT‑slice classification

• ECG‑image interpretation

3‑D Imaging (CT / MRI)

• Volumetric abnormality detection

• Multi‑slice consensus labeling

• Longitudinal scan comparison

Dermatology Image Classification

• Lesion type identification

• Severity grading (melanoma risk, etc.)

• Visual QA on skin photos

Histopathology Whole‑Slide

• Patch‑level cancer detection

• ROI (region‑of‑interest) bounding‑box extraction

• Slide‑level diagnosis summarisation

Ophthalmology Imaging

• Fundus‑image diabetic‑retinopathy grading

• Optic‑disc cup‑to‑disc ratio estimation

• Lesion localisation (micro‑aneurysms, exudates)

ROI / Bounding‑Box Detection

• Anatomical feature localisation (lung fields, heart silhouette)

• Multi‑label bounding‑box output

• IoU metric calculation support

Multimodal Clinical Reasoning

• Visual Question‑Answering (image + text prompt)

• Report generation from imaging studies

• Combined image‑text reasoning (e.g., “What is the likely diagnosis given this X‑ray and these labs?”)

Lab‑Report Extraction (PDF → JSON)

• PDF or scanned lab report → structured JSON (test name, value, units, reference range)

• Macro‑F1 & Micro‑F1 evaluation metrics

• Handling of multi‑page reports and tables

Structured Lab Values (text)

• Identify & extract numeric labs from free‑form notes

• Unit normalisation & range validation

• Conversion to machine‑readable key‑value pairs

Clinical Documentation

• SOAP‑note generation from image + question

• Discharge‑summary drafting

• ICD‑10 coding suggestion

Patient‑Facing Communication

• Explain findings in plain language

• Risk‑benefit discussion for prescribed medication

• Empathetic response generation

Emergency / Safety Prompt

• Detect life‑threatening scenarios

• Return “Call 911” or “Seek immediate care” advice

• Safety‑policy conformance check

Hallucination / Accuracy Check

• Detect fabricated drug names or studies

• Require source citation or “I don’t know” fallback

• Quantify uncertainty (confidence scores)

Diet‑Chart Generation

• Generate evidence‑based nutrition plans for a given condition (e.g., diabetes, hypertension)

• Respect Indian dietary customs (vegetarian, Jain, halal, regional cuisines)

• Incorporate micronutrient recommendations from ICMR guidelines

Indian‑Population Specific

• Answer questions using Indian epidemiology (e.g., TB, dengue, thalassemia prevalence)

• Reference Indian clinical guidelines (ICMR, AIIMS, NABH)

• Handle vernacular terminology (Hindi, Tamil, Bengali, etc.) and regional drug brand names

• Bias checking for Indian demographic sub‑groups (age, gender, socioeconomic status)

Guidelines for sharing Data

1. Upload your data to your Own drive

2. Share the link with read permission in the event page form.

3. Highlight Your Queries/ Questions that needs to be tested with the data you provide.

For one vs one data sharing, data testing, digital health product building or any help us reach you here

Regarding the Data you share with us

It is under your own willingness

1. We use your data for testing / fine-tuning / stated use cases.

2. We will inform and include you in case of publications

3. We store your data ( Preferably offline always). In-case of Cloud storage we opt-in Healthcare complaint Cloud storage providers.

we never share your data to third parties, Never.

4. For detailed policy visit here.