Aims: The objective of this study is to assess the long-term incidence of contralateral second breast cancer in patients who have undergone breast cancer radiotherapy. Artificial intelligence and advanced analytics are employed to establish the correlation between radiation dose distributions and the risk of developing a second cancer.
Methods: Approximately 40,000 patients received radiotherapy for invasive breast cancer in Queensland over the past two decades. Among these patients, 1,448 individuals were identified from the Queensland Oncology Repository who were subsequently diagnosed with a second cancer at a different site. In this preliminary study, we collected DICOM image and RT datasets from a single institution for patients who had developed contralateral breast cancers following radiotherapy. The dose to the contralateral breast was compared against patients who had received radiotherapy but remained free from contralateral breast cancer. Deep learning autosegmentation models were generated to segment organs at risk (OAR), including the right and left breasts, right and left lungs, and heart. Autosegmentation was performed on all datasets missing contours for these structures.
Results: Preliminary findings indicate that patients who subsequently developed contralateral breast cancer received a mean dose of 0.68 ± 0.62 Gy to the contralateral breast, while those who did not develop contralateral breast cancer received a mean dose of 0.33 Gy ± 0.24 Gy. Dice similarity coefficients for segmentations generated by the model for the left breast, right breast, heart, left lung, and right lung were calculated as 0.91, 0.89, 0.92, 0.97, and 0.97, respectively.
Conclusion: This is an ongoing study and is being extended to collaborate with other institutions. It will enable a more precise assessment of the treatment-related risk factors for radiation-induced breast cancers and this information can be used to optimise existing treatment techniques.