Team RadioGyan

572 posts

Team RadioGyan

Team RadioGyan

@TeamRadioGyan

Official Account of Team RadioGyan. New radiology learning posts for #radres everyday. #FOAMrad

More radiology resources 👉 Katılım Haziran 2021
1 Takip Edilen2.3K Takipçiler
Dr. Ameya Kawthalkar
Dr. Ameya Kawthalkar@ASK_MSK·
How I built a radiology AI model with zero coding experience: I am a complete noob at coding and AI. 6 months ago I did not know what Google Colab was. The only AI I have used in clinical practice was to translate MRI and ultrasound reports into French and Spanish for athletes at the 2024 Paris Olympics as a musculoskeletal radiologist. The only AI course I have attended is the Royal College of Radiologists' Blended Radiology AI learning course. But thanks to large language models (LLMs), in one intense month I was able to train and deploy an AI model for paediatric bone age estimation using hand X-rays. Bone age estimation is important for planning treatment for patients with endocrine and growth disorders. Manual bone age estimation by a radiologist takes 10-15 min per Xray with mean absolute error (MAE) of 9 to 12 months. Bone Age Alpha, the AI model I managed to train, does it in 10-15 seconds per Xray with MAE of 6 to 7 months. ChatGPT 4.0 and Grok 3 taught me step by step how to build the AI model, the backend and frontend. They also explained the code and any AI concepts I wanted to learn. They even taught me how to segment bones on X-rays in ITK-SNAP. For designing the implementation spec, actual code generation and debugging I used o3-mini-high. I also built the FastAPI backend with o3-mini-high in a day on Hugging Face Spaces. Finally I built the frontend with React in Replit with Replit Agent and Replit Assistant (powered by Claude 3.7 Sonnet) which was the shortest step and finished in 2 hours. After much trial, error, frustration, running into hundreds of walls, finding a thousand ways which did not work and starting from scratch multiple times, I finally managed to build this model: Trained on the RSNA bone age dataset in Google Colab using Nvidia's A100 GPU. Triple model setup: 1. DeepLabV3+ for automated segmentation 2. EfficientNetV2M for bone age analysis 3. A trained linear regression model for final calibration and refinement Also put in smartphone camera integration, for potential role in resource poor settings around the world with film screen radiography. Bone Age Alpha will always be free and open source. I will always keep any radiology and healthcare AI and apps I build open source and free to use. You now have the 95th to 97th percentile level elite coders in your pocket with o3, Grok and Claude. They are the best teachers one could ever want. If I can build this, any doctor can build AI models and apps now, which help patients and other doctors. When more doctors working in the trenches in clinical practice become involved in building healthcare AI, more AI will be developed which actually makes a clinical impact. Lastly the most important point: 🚨Bone Age Alpha is for educational purposes only. Clinical deployment requires rigorous clinical trials of AI models and regulatory approvals. Link to the web app, details of training and GitHub repo in 🧵 below
English
18
22
114
19.4K
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
First Trimester SRU Lexicon. Read the full article by Rodgers et al. in Radiology. Follow us to get more radiology content in your feed. #RGpearls #radiogyan #radiology
Team RadioGyan tweet media
English
0
0
1
130
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
Interventional Radiologists: Making waves in and out of the hospital! Check out more radiology memes: bit.ly/4eeD1dI
English
0
0
0
353
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
When ‘Do Not Disturb’ mode gets a whole new meaning in radiology. Check out more radiology memes: bit.ly/4eeD1dI
Team RadioGyan tweet media
English
0
0
0
254
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
Radiologists: The only species known to be photophobic in broad daylight! Check out more radiology memes: bit.ly/4eeD1dI
English
0
0
0
173
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
Radiology Residents: Masters of the ‘It could be anything’ school of thought! Check out more radiology memes: bit.ly/4eeD1dI
English
0
0
0
227
Team RadioGyan
Team RadioGyan@TeamRadioGyan·
Wondering what are the best books for radiology? Check out this comprehensive list : bit.ly/4e3pPIJ
Team RadioGyan tweet media
English
0
1
5
300