Feasibility of anomaly score detected with deep learning in irradiated breast with reconstruction
Dong-Yun Kim,
Korea Republic of
PO-1215
Abstract
Feasibility of anomaly score detected with deep learning in irradiated breast with reconstruction
Authors: Dong-Yun Kim1, Soo Jin Lee2, Eun Kyu Kim3, Eunyoung Kang3, Yujin Myung4, Chan Yeong Heo4, In Ah Kim5, Bum-Sup Jang5
1Seoul National University Hospital , Radiation Oncology , Seoul, Korea Republic of; 2Seoul National University , College of Medicine , Seoul, Korea Republic of; 3Seoul National University Bundang Hospital , Surgery, Seongnam, Korea Republic of; 4Seoul National University Bundang Hospital , Plastic and Reconstructive Surgery, Seongnam, Korea Republic of; 5Seoul National University Bundang Hospital , Radiation Oncology , Seongnam, Korea Republic of
Show Affiliations
Hide Affiliations
Purpose or Objective
The aim of this study is to evaluate cosmetic outcomes of the reconstructed breast in breast cancer patients using anomaly score (AS) detected by generative adversarial network (GAN) deep learning algorithm.
Material and Methods
A total of 251 normal breast images from patients who underwent breast-conserving surgery were used for training anomaly GAN network. GAN-based anomaly detection was used to calculate abnormalities as an AS with Z-score standardization. Then, we retrospectively reviewed 61 breast cancer patients who underwent mastectomy followed by breast reconstruction with autologous tissue or tissue expander. All patients were treated with adjuvant radiation therapy (RT) after breast reconstruction and computed tomography (CT) was performed three time points; before RT (Pre-RT), one year after RT (Post-1Y), and two years after RT (Post-2Y).
Results
Compared to Pre-RT, Post-1Y and Post-2Y demonstrated higher AS, indicating more abnormal cosmetic outcomes on paired-T test (Pre-RT vs. Post-1Y, P=0.015 and Pre-RT vs. Post-2Y, P=0.011). Pre-RT AS was significantly higher in patients having major breast complications (P=0.016). Autologous reconstruction showed better AS than tissue expander insertion at pre-RT (2.00 vs. 4.19, P=0.008) and post-2Y (2.89 vs. 5.00, P=0.010). Linear mixed effect model revealed that days after baseline were significantly associated with increasing AS (P=0.007). Also, the use of tissue expander was associated with steeper rise of AS, compared with autologous tissue (P=0.015). Fractionation regimen (conventional fractionation vs. hypofractionation) was not associated with the change of AS (P=0.389).
Conclusion
Anomaly score detected by deep learning might be feasible in predicting cosmetic outcomes of RT-treated patients with breast reconstruction. AS should be validated in clinical trial settings.