Fig. 2: Comparison of the predictive ability for adverse pathology between the NAFNet-classifier and the ResNet50-classifier. | npj Precision Oncology

Fig. 2: Comparison of the predictive ability for adverse pathology between the NAFNet-classifier and the ResNet50-classifier.

From: A deep learning model, NAFNet, predicts adverse pathology and recurrence in prostate cancer using MRIs

Fig. 2

a NAFNet-classifier and ResNet50-classifier prediction visualized results on the MRI image depicting the probability of adverse pathology in identified regions. The left panels show MRI images from a patient who did not experience any adverse pathology events. The NAFNet-classifier predicted an AP probability of 0.126, while the ResNet50-classifier predicted a probability of 0.151. While the right panels depict MRI images from a patient with an adverse pathology event. The NAFNet-classifier gave a 0.632 AP probability, and the ResNet50-classifier had a probability of 0.264. The colour scale on the right represents the probability of adverse pathology events in each pixel within the identified region, with red showing a higher probability; and blue showing a lower probability. b Receiver operating characteristics curves and (c) decision curve analyses of NAFNet-classifier and ResNet50-classifier in predicting adverse pathology events based on external test set. Abbreviations: AP adverse pathology, ADC apparent diffusion coefficient T2WI, DWI diffusion-weighted imaging, T2-weighted magnetic resonance imaging, AUC area under the receiver operating characteristic curve, CI confidence interval.

Back to article page