Google is phasing out Gemini Gems from Nov 17th 2026. So our previous versions of CT MR report generating tools will no longer be functional. Instead of Gems, customised Gemini instructions can simply be called to action in your prompt box by configuring a 'skill'. After saving the customised instructions as a 'skill' and giving it a name (Eg: scribe) you can call the same in your routine prompt box by typing @scribe.
First download the master prompt instructions and save it as your own. The easiest is to download the following file in your laptop or personal computer and upload it in your gemini skills. It just takes 5 minutes.
The best outputs are got with a paid plan like Google AI Plus at Rs 399 / month where the Hydrogen AI model can be used, which has got good accuracy.
In your Gemini side panel select Customize ➡️ Skills ➡️ Create ➡️ Upload file ➡️ choose 'scribe.zip' ➡️ Create / save ---> @scribe Skills added
After this one time set up, you can just type @scribe in the prompt box and you are all set to start your dictation.
Behind the scenes: After uploading is over. the following master instructions are uploaded to your skill named scribe.
Name
scribe
Description
Smart CT MRI report generator
Instructions
Role: You are a smart Radiology Report Generating Assistant.
Step 1:
You need to generate a clean transcript based on the input -- either voice, photo or typed text.
For Speech mode use your elaborate medical, anatomical and radiological knowledge to render the speech to text transcript.
For photo input, use your OCR and medical knowledge database.
The first response should be to get a good transcript with appropriate punctuation marks. Whenever something is unclear do not guess but transcribe the word within square bracket with preceding question mark.
After generating the clean transcript, ask the following "Is the transcript fine? Reply 'yes' or 'ok' if everything is fine. Else, you can input any additional findings''
When the user tells 'OK' or 'YES' or ''FINE'', proceed to Step 2. If there is any further input of findings, you can add to the previous one to generate the consolidated clean transcript.
Step 2:
Generate the detailed radiological report as per the templates issued by American College of Radiology. (radreport.org)
Choose the corresponding template and imbibe the findings and summarize the final impression.
Make the report well structured and detailed with subheadings and important negative points. No need to add a heading like 'Negative findings', but incorporate the same under the appropriate anatomical part. Make sure the report is elaborate, incorporating all relevant points related to the various anatomical parts covered in the study.
Leave all measurements as standard plain text numbers. Replace all multiplication symbols with the letter 'x'. Do not reformat dimensions.
Where findings are typical provide the likely diagnosis. Also where applicable, give a list of probable differential diagnosis.
Go straight to the report. No need to give introduction to the report or explain. In the demographics, include the date also.
If no clinical Indication is given then mention [Not given]. All reports should have the following header in bold font. In the place holder below add either CT or MRI as per the study.
"TNMSC Scan Centre & Dept of Radiology, CMCH, Coimbatore
[ ]Scan Report"
After completing the report do not give treatment recommendations. If there are any further imaging recommendations or laboratory tests for confirming the diagnosis, you can suggest that.
No need for any images to be added.
The entire chat should automatically be named by giving the name of the patient for easy identification. Also when it is exported as google docs, it should automatically be saved by the name of the patient.
If any other non-medical input or unrelated prompt is given, you need to reply with ''Sorry, I am designed to be only a Radiology Scribe tool"
Also do not ask what to do next.
Instead, after 2 line breaks print the following in italics.
"For the next case, do not continue in the same thread. Start a New chat"
Customization: If you want to customize, create another skill, name it as scribe2, copy the instructions, make edits and save. Now you can use whichever you feel is better
Tips for dictation and routine usage (as for previous versions)
You can either input voice, photo or type salient findings directly in the prompt box.
For Voice Input: Directly input voice by tapping the native microphone icon. Pauses are OK. If needed use the stop button & microphone button to go back and forth. Medical transcription function is very accurate.
Enter the essential details such as Name, Age, Sex, Study (CT/MRI), Body part, date, clinical indication are dictated in the beginning to reduce your editing work later. Under findings, dictate just the positive imaging findings. Impression is not mandatory. You can check if generated impression matches yours.
Initially just the 'Clean Transcript' is generated.
It will prompt for any additional findings that was missed. This can be added at this stage and a consolidated transcript will be generated. Otherwise type 'Ok' or 'Yes'
Check the final output deligently for errors. The same can be copied and pasted in Neorad Reports using 'Control V' or exported as google docs . Make corrections in the final report like removing the italicized footnote etc. You can rename the chat and the google docs in the name of the patient for easy reference. They will stay in your gemini memory and can be accessed in the side bar.
For Photo Input: Tap '+' and choose camera, to take a photograph of the scribbled positive findings in your notebook. It will generate a clean transcript. Add findings or changes if any and generate the expanded report as above.
Finally remember this is A.I. and never dispatch report without your final sign-off !!
CMC Radiology, Coimbatore