The Double Dilemma in Multi-Task Radiology Report Generation: A Gradient Dynamics Analysis and Solution

Researchers proposed CAME-Grad, a plug-and-play optimizer for multi-task radiology report generation.

The paper explains why linear scalarization can fail, framing the problem as gradient dynamics in training.

Across eight methods on MIMIC-CXR and IU X-Ray, CAME-Grad improved clinical performance consistently.

The work targets a common tradeoff in medical AI between report quality and clinical supervision.