Objective: Invasion can be detected in up to 50% of excision specimens in cases diagnosed with ductal carcinoma in-situ (DCIS) in tru-cut biopsies. The predicting of invasion is critical for managing the treatment. The objective of this study was to ascertain the predictive role of histological characteristics and estrogen receptor (ER) and HER2 expression of DCIS with respect to invasion and to create an artificial intelligence (AI)-assisted invasion risk scoring.

Material and Methods: A total of 140 tru-cut biopsies with DCIS were included. We evaluated the predominant structural pattern, nuclear grade, presence of comedonecrosis, and ER/HER2 expressions and their relationship with invasion upstage. Based on the results, an invasion risk scoring system was created using OpenAI-ChatGPT 5.2.

Results: Of the 140 cases, 70 (50%) were high-grade (DCIS-3), 49 (35%) were intermediate-grade (DCIS-2), and 21 (15%) were low-grade (DCIS-1). DCIS-1 and DCIS-2 showed significant association with a cribriform pattern, and DCIS-3 showed with a solid pattern. When the results were compared with those of the final specimens, invasion upstaging occurred in 9.5% of DCIS-1, 38.7% of DCIS-2, and 52.8% of DCIS-3 cases. A solid pattern, high nuclear grade, comedonecrosis, and ER-/HER2- profile showed high correlation with invasion. In AI-assisted risk scoring, DCIS cases were divided into three groups: low, medium, and high risk. The invasion upstaging rates in these groups were 11.44%, 28.97%, and 64.19%, respectively.

Conclusion: DCIS with high nuclear grade, comedonecrosis, a solid pattern, and the ER-/HER2+ or ER-/HER2- immunoprofile should be considered high risk for invasion. They should be approached as invasive tumors and evaluated in terms of lymph node staging and neoadjuvant treatment.