Go to MIDL 2026 Conference homepagePhysician-Guided Learning with Attention Mechanism for Non-Tuberculosis Mycobacterium Disease Classification Using 3D Chest CT Images
Keywords: Computed Tomography, Human-in-the-Loop, Expert-Guided Learning, Attention Mechanism
Abstract: The growing demand for interpreting Computed Tomography (CT) scans has driven the adoption of deep learning techniques, particularly Convolutional Neural Networks (CNNs), to assist physicians in managing their increasing workload. However, CNNs typically require large, well-annotated datasets to learn clinically meaningful imaging patterns, and such datasets remain limited due to the volumetric and resource-intensive nature of CT imaging. Incorporating physician expertise into the training process offers a promising path to overcoming this data limitation. In this study, we introduce a Human-in-the-Loop (HITL) workflow that enables physicians to directly guide CNN training by (1) providing feedback on model-generated attention maps and (2) validating or correcting predicted labels. This interactive process integrates domain knowledge into both representation learning and model interpretability. We evaluate the proposed HITL workflow on a chest CT dataset of patients with Non-tuberculous Mycobacteria (NTM) infection, a clinically challenging pulmonary disease that often requires long-term imaging follow-up to assess disease progression. Experimental results demonstrate that the HITL approach improves disease progression classification and produces clinically meaningful attention patterns, highlighting the value of physician-guided learning in medical imaging AI.
Primary Subject Area: Safe and Trustworthy Learning-assisted Solutions for Medical Imaging
Secondary Subject Area: Integration of Imaging and Clinical Data
Registration Requirement: Yes
Visa & Travel: No
Read CFP & Author Instructions: Yes
Originality Policy: Yes
Single-blind & Not Under Review Elsewhere: Yes
LLM Policy: Yes
Submission Number: 272
Loading