Our recent study analyzed a subset of 5,136 mammograms, due to data constraints, from a decade-long screening program (five rounds) involving over 22,000 mammograms from around 9,000 women. Our objective was to evaluate the AI’s detection capabilities and assess various reading scenarios combining AI with human reading. The results highlighted the AI system’s performance with an AUC of 90%.
However, interpreting AI-assigned risk scores presents challenges, necessitating the establishment of optimal thresholds to enhance effectiveness. Setting an appropriate threshold is crucial to balance cancer detection and reduce false negatives. Our calculated threshold yielded a balanced sensitivity of 72% and a high specificity of 93%, effectively detecting both screening-detected and interval cancers. The AI system identified over 80% of screening-detected cancers and more than 50% of interval and missed cancers. It also demonstrated the potential to detect one-fifth of cancers in one or two prior screenings, highlighting its ability to decrease interval cancers.
We explored various reading scenarios, including using AI as a second reader and triage tool. The triage tool scenario was the most effective for detecting interval cancers and minimizing missed cancers. Combining human reading with AI, particularly in a triage capacity, emerged as the optimal approach, reducing the workload by almost 70% and increasing accuracy by 30% when combined with a single human reader.
Implementing AI in breast cancer screening programs significantly reduces radiologists’ workload and decreases interval cancers, leading to earlier diagnoses and better patient outcomes. We believe that adopting AI as a triage mechanism in screening programs is a transformative step towards more efficient and accurate cancer detection.
Key points:
- Incorporating AI as a triage tool in screening workflow improves sensitivity (72.38%) and specificity (92.86%), enhancing detection rates for interval and missed cancers.
- AI-assisted triaging is effective in differentiating low and high-risk cases, reduces radiologist workload, and potentially enables broader screening coverage.
- AI has the potential to facilitate early diagnosis compared to human reading.
Article: Diagnostic capabilities of artificial intelligence as an additional reader in a breast cancer screening program
Authors: Mustafa Ege Seker, Yilmaz Onat Koyluoglu, Ayse Nilufer Ozaydin, Sibel Ozkan Gurdal, Beyza Ozcinar, Neslihan Cabioglu, Vahit Ozmen & Erkin Aribal
