ItpCtrl-AI: End-to-end interpretable and controllable artificial intelligence by modeling radiologists' intentions

Trong-Thang Pham, Jacob Brecheisen, Carol C. Wu, Hien Van Nguyen, Zhigang Deng, Donald A. Adjeroh, Gianfranco Doretto, Arabinda Choudhary, Ngan Le

Abstract: Highlights•We propose ItpCtrl-AI, a controllable interpretable method for chest X-ray diagnosis.•ItpCtrl-AI generates gaze heatmaps to aid in diagnosis and allows user control.•We introduce Diagnosed-Gaze++ dataset, aligning medical findings with eye gaze data.•Experiments show the method effectively identifies findings and reproduces gaze patterns.