Transfer Learning
Fine-tuning CAD4TBv7 with only 525 paediatric CXRs improved AUC from 0.58 to 0.72 against a radiological reference standard, a statistically significant gain. Transfer learning — fine-tuning adult-trained algorithms on smaller paediatric d…
1 sources - 4 claims
Fine-tuning CAD4TBv7 with only 525 paediatric CXRs improved AUC from 0.58 to 0.72 against a radiological reference standard, a statistically significant gain. Transfer learning — fine-tuning adult-trained algorithms on smaller paediatric datasets — is a feasible alternative to building paediatric CAD from scratch. Building a paediatric CAD system from scratch would require prohibitively large labelled image datasets. Approximately two-thirds of the CAPTURE repository will be made available as a training set for developers to fine-tune paediatric-specific algorithms.