From healthcare.hospital import datascience
summary
Project applies anomaly detection to organ contours in head and neck radiotherapy, claiming a pragmatic isolation forest + feature engineering approach (concavity, symmetry, perimeter/area) that runs in parallel with clinical workflow and surfaces imaging errors, contouring disagreements and anatomy changes. Pipeline uses shapely features and a bespoke RDF database → SQL extraction because DICOM is graphy and cloud is restricted. Result: useful flags and explainable feature importances, but labeling, infrastructure and GDPR limit wider deployment.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:06Team and project
- 2:31Radiotherapy workflow
- 4:58Research vs clinical reality
- 7:01Medical imaging standards
- 8:25Tumor delineation focus
- 9:27Head and neck cancer
- 10:50Organ contouring process
- 11:33Anomaly detection in contours
- 15:26Isolation Forest
- 16:29Labeling challenges
- 17:34Hybrid approach
- 18:37Explaining anomalies
- 20:04Data science pipelines
- 21:50Data infrastructure limits
- 22:55Visualization and engagement
- 24:02Results and insights
- 26:35Final remarks
- 28:00Q&A LIME explainers
- 29:47Q&A global collaboration
- 31:40Q&A medical applications
- 33:25Q&A patient-reported data
- 34:10Q&A clinical roles
- 35:57Q&A institutional context
- 37:54Q&A regulatory barriers
related talks
11mInterview
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
03mAI solutions for healthcare
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
25mComputer Vision in Melanoma Detection
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
50mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
36mBeyond Medical Image Segmentation. The Road Towards Clinical Insights