The integration of artificial intelligence (AI) into Diagnostic Radiology is both promising and uncertain. In Radiology Department of Prince of Wales Hospital (PWH), we piloted a vendor AI solution, focusing on two modules: lung nodule detection in CT thorax and pulmonary embolism (PE) detection in CT pulmonary angiogram (CTPA). This presentation shares our experience, emphasizing AI's potential to enhance diagnostic workflows while addressing real-world challenges.
For lung nodule detection, internal validation compared AI results with radiologists' findings, revealing promising results in nodule detection. This suggests AI's utility in flagging cases requiring early attention. However, workflow efficiency comparisons (radiologist-only vs. radiologist + AI) showed mixed outcomes: while AI reduced initial screening time for some radiologists, its impact varied across individuals, underscoring the need for workflow optimization.
In PE detection, the AI demonstrated reliable performance in detecting acute emboli from the pulmonary trunk to lobar branches. This capability has the potential to expedite critical alerts to clinicians, shortening time-to-diagnosis for life-threatening PE and enabling faster clinical management, which may finally improve patient outcomes.
Radiographers serve as frontline decision-makers, tasked with identifying cases requiring urgent or earlier radiological reporting. By integrating AI into routine workflows, this empowers radiographers to swiftly pinpoint cases requiring prioritized reporting amidst high patient volumes, leveraging AI-generated findings to shortlist cases with significant findings (e.g., suspicious nodules) or acute conditions (e.g., PE) for expedited radiologists' review. This synergy bridges AI's analytical power with actionable clinical care.
While AI adoption at PWH remains a pilot initiative, our findings underscore its dual role as a supplementary diagnostic tool and workflow accelerator. Challenges such as inconsistent efficiency gains persist, but the technology shows promise in enhancing diagnostic confidence and prioritizing time-sensitive cases.
