Purpose: Recurrent nasopharyngeal carcinoma after treatment can be difficult to recognize as there are overlapping imaging findings with post-radiotherapy changes. We have established an intelligent computerized system that differentiates recurrent carcinoma and normal condition using artificial neural network based on studies of Positron Emission Tomography/Computed Tomography with limited cases (n=16). Method: Twenty-one validated radiological features were used as inputs for the neural network. The neural network was trained by leave-one-out cross-validation (LOOCV) method with 2 outputs. Results: Our observer study indicated that when the radiologists are provided with the ANN output they make better decisions and there was significant improvement for junior radiologists after using neural network results (p<0.01). Conclusion: This study demonstrated the feasibility of using artificial neural network for deciphering complex imaging pattern of recurrent nasopharyngeal carcinoma after treatment with limited case samples and improved the diagnostic accuracy of PET/CT for recurrent nasopharyngeal carcinoma.
|Number of pages||5|
|Journal||International Journal of Radiology & Medical Imaging|
|Publication status||Published - Feb 2017|
Tang, F., Pang, E. YN., & Chan, T. (2017). A diagnostic model for recognition of recurrent nasopharyngeal carcinoma in Positron Emission Tomography/ Computed Tomography (PET/CT) based on artificial neural networks. International Journal of Radiology & Medical Imaging, 3(117), 1-5. https://doi.org/10.15344/2456-446X/2017/117