The aim of this page is to organize information for machine learning in medicine (MLinMed) research by (1) describing the AIGIP montly videocast, (2) introducing tools for MLinMed research, and (3) listing Youtube channels and websites about MLinMed.
AIGIP videocast is a monthly 40-minute video series hosted on YouTube and LinkedIn, dedicated to summarizing recent articles showcasing the latest advancements in artificial intelligence applied to gastroenterology and pancreatology. Each episode features a deep dive into one selected paper, subject to rigorous code review to ensure AI reproducibility. On our platform, you can also discover more about our dedicated team, any potential conflicts of interest, supplementary information, and our approach to curating and presenting these insightful articles.
Date: 21 April 2023
Team: Amir Safavi, Arian Salahi, Ghazaleh Sadeghi, Mahsa Aghamohamadpour
Conflict of Interest: AS received compensation from MEDICAI company in 2022 as an R&D associate.
In the first episode of AIGI, we discuss:
1- [Intro] Introducing the aim of AIGI, and how we select papers
1- [AI-Hot topic] The letter to stop expert language models such as Chat-GPT4
3- [AI- Learn] sources to learn basic concepts of ML
Checklist for Reporting MLinMed
Where your data come from?
- Define your context
- Define time and people involved in data gathering
- Basic characteristics of dataset
- How much missing data we have? What did you do with missing data?
- Method for feature selection (expert opinion or statistics or both)
How good is our model prediction?
- Confusion Matrixxxxxxxxxxxx (super important)
- Classification: F1-score (Precision, Recall=Sensitivity, Specificity)
- Probabilities: AU-ROC or PR AUC (percision-recall)
- Calibration Curve
How much data we need?
- Performance at varying training set size
Why our model make mistakes?
- Sensitivity analysis
- Reporting performance in each subgroup (such as confusion matrix)
Time and GPU/CPU needed for your model?
- GPU/CPU for training
- GPU/CPU and time for prediction of one new data
How interpretable your model is?
- Use explainable AI (XAI) tools in your model
Checklist for quality assessment of studies
3D Slicer
ITK Snap
Qupath
Youtube channels, newsletter, and websites for MLinMed
1- Doctor Penguin newsletter: Receive a weekly summary and discussion of the top AI + Healthcare research papers.
