Diagnostic and Interventional Radiology
Artificial Intelligence and Informatics - Original Article

Application of ultrasonic dual-mode artificially intelligent architecture in assisting radiologists with different diagnostic levels on breast masses classification

1.

Department of Ultrasound, Shanghai Jiao Tong University School of Medicine, Shanghai General Hospital Shanghai, China

2.

Shanghai Key Lab of Digital Media Processing and Transmission, Shanghai Jiao Tong University, Shanghai, China

3.

Department of Mathematics and Computer Science, Centre for Analysis, Scientific Computing, and Applications W&I, Eindhoven University of Technology, Eindhoven, Netherlands

Diagn Interv Radiol 2020; 1: -
Read: 278 Published: 27 May 2020

PURPOSE: We aim to compare the diagnostic performance and inter-observer variability of different radiologists in breast tumor classification with or without the aid of an innovative Dual-Mode artificial intelligence (AI) architecture which could automatically integrates information from US-mode and SWE-mode.

METHODS: Diagnostic performance assessment was performed with a test subset, containing 599 images (from September 2018 to February 2019) from 91 patients including 64 benign and 27 malignant breast tumors. Six radiologists (three inexperienced radiologists and three experienced radiologists) were assigned to read images independently and then make secondary diagnosis with knowledge of AI results (Independent-Diagnosis mode and Secondary-Diagnosis mode). Sensitivity, specificity, accuracy, receiver-operator characteristics (ROC) curve analysis and Cohen's k statistics were finally calculated.

RESULTS: As for the inexperienced radiologists’ group, from Independent-Diagnosis mode to Secondary-Diagnosis mode, the average area under the ROC curve (AUC) of US-mode increased from 0.722 to 0.765 (P = 0.050) and from 0.794 to 0.834(P = 0.019) for Dual-Mode. The average AUC of experienced radiologists’ group was significantly higher with AI system on US-Mode (0.812 vs 0.833, P = 0.039), but not for Dual-Mode (0.858 vs 0.866, P = 0.458). At the Secondary-Diagnosis mode, the better inter-observer agreement for all radiologists was obtained on US-Mode (P = 0.003) from fair agreement to moderate agreement. On Dual-Mode, substantial agreement was seen among the experienced radiologists (0.65 to 0.74, P = 0.017) and all the radiologists (0.62 to 0.73, P = 0.001).

CONCLUSION: The diagnostic performance improvement is more distinguished for the inexperienced radiologists with AI assistance, meanwhile, the experienced radiologists benefit more from AI in reducing inter-observer variability.

EISSN 1305-3612