초록접수 현황
| 20F-120 | 구연 발표 |
Automated Detection Improves Identification of Oligometastatic Disease and Prevention of Missing Metastases during Metastasis-directed Therapy
Jason Joon Bock Lee1, Jee Suk Chang1, Byung Min Lee1, Jin Sung Kim1, Yongjin Chang4, Young Joo Suh2, Seong Yong Park3
Departments of 1Radiation Oncology, 2Radiology and 3Thoracic and Cardiovascular Surgery, Yonsei University College of Medicine, Seoul, Korea
4Coreline Soft, Co., Ltd, Seoul, Korea
Purpose : To evaluate the clinical utility of computer-aided detection (CAD) in the identification of lung nodules in pulmonary oligo-metastases.
Methods : A total number of suspicious lung lesions were counted from the CT datasets of 311 patients with lung metastases from colorectal cancer by radiation oncologist (non-expert), Aview Lung Nodule computer-aided detection (CAD) software, and CAD-aided radiation oncologist. Predictive ability according to the number of lung lesions were then compared in survival models. In auxiliary analysis, a radiology expert blindly reviewed 50 randomly selected patients with all serial CT scans to estimate the possible true lesions from the suspicious lesions.
Results : CAD detected 4147 suspicious lesions. 632 lesions in 212 patients were removed and 163 lesions in 45 patients were added after the review process. Patients were divided into five groups (1, 2, 3, 4, and ≥5 lesions) according to the number of identified lesions. In survival prediction model, CAD-aided approach showed the highest predictive performance with a time-dependent AUC of 0.84 at 5-year and good calibration, followed by manual approach and CAD only approach. 5-year survival for CAD-aided respective groups were 75%, 53%, 46%, 30%, and 23%. In auxiliary analysis of patients with 1-3 lung lesions, radiologist detected 38 suspicious lesions (compared to 23, 58 and 39 lesions in non-expert, CAD and CAD-aided non-expert, respectively) and 63% of them were interpreted as true lesions.
Conclusion : Utilizing CAD can improve the identification of oligo-metastatic state and avoid potential problems of missing small metastasis lesions during localized therapy.
Methods : A total number of suspicious lung lesions were counted from the CT datasets of 311 patients with lung metastases from colorectal cancer by radiation oncologist (non-expert), Aview Lung Nodule computer-aided detection (CAD) software, and CAD-aided radiation oncologist. Predictive ability according to the number of lung lesions were then compared in survival models. In auxiliary analysis, a radiology expert blindly reviewed 50 randomly selected patients with all serial CT scans to estimate the possible true lesions from the suspicious lesions.
Results : CAD detected 4147 suspicious lesions. 632 lesions in 212 patients were removed and 163 lesions in 45 patients were added after the review process. Patients were divided into five groups (1, 2, 3, 4, and ≥5 lesions) according to the number of identified lesions. In survival prediction model, CAD-aided approach showed the highest predictive performance with a time-dependent AUC of 0.84 at 5-year and good calibration, followed by manual approach and CAD only approach. 5-year survival for CAD-aided respective groups were 75%, 53%, 46%, 30%, and 23%. In auxiliary analysis of patients with 1-3 lung lesions, radiologist detected 38 suspicious lesions (compared to 23, 58 and 39 lesions in non-expert, CAD and CAD-aided non-expert, respectively) and 63% of them were interpreted as true lesions.
Conclusion : Utilizing CAD can improve the identification of oligo-metastatic state and avoid potential problems of missing small metastasis lesions during localized therapy.
책임저자: Seong Yong Park
Department of Thoracic and Cardiovascular Surgery, Yonsei University College of Medicine
발표자: Jason Joon Bock Lee, E-mail : jbwolf86@yuhs.ac