Deep learning classification of inverted papilloma malignant transformation using 3D convolutional neural networks and magnetic resonance imaging

Abstract

Background: Distinguishing benign inverted papilloma (IP) tumors from those that have undergone malignant transformation to squamous cell carcinoma (IP-SCC) is important but challenging to do preoperatively. Magnetic resonance imaging (MRI) can help differentiate these 2 entities, but no established method exists that can automatically synthesize all potentially relevant MRI image features to distinguish IP and IP-SCC. We explored a deep learning approach, using 3-dimensional convolutional neural networks (CNNs), to address this challenge.Methods: Retrospective chart reviews were performed at 2 institutions to create a data set of preoperative MRIs with corresponding surgical pathology reports.The MRI data set included all available MRI sequences in the axial plane, which were used to train, validate, and test 3 CNN models. Saliency maps were generated to visualize areas of MRIs with greatest influence on predictions. Results: A total of 90 patients with IP (n = 64) or IP-SCC (n = 26) tumors were identified, with a total of 446 images of distinct MRI sequences for IP (n = 329) or IP-SCC (n = 117). The best CNN model, All-Net, demonstrated a sensitivity of 66.7%, specificity of 81.5%, overall accuracy of 77.9%, and receiver-operating characteristic area under the curve of 0.80 (95% confidence interval, 0.682-0.898) for test classification performance. The other 2 models, Small-All-Net and Elastic-All-Net, showed similar performance levels. Conclusion:A deep learning approach with 3-dimensional CNNs can distinguish IP and IP-SCC with moderate test classification performance. Although CNNs demonstrate promise to enhance the prediction of IP-SCC using MRIs, more data are needed before they can reach the predictive value already established by human MRI evaluation.

Supervised Contrastive Learning to Classify Paranasal Anomalies in the Maxillary Sinus

“…We also show that a combinination of cross-entropy loss and supervised contrative loss improve the performance of the model. Finally, compared to other works that use deep learning for paranasal inflammation study [8,10,11], ours is the first work that tries to achieve label efficiency and thus attempts to reduce the workload of physicians. A limitation of our work is that the classification accuracy is still not satisfactory.…”

“…In conclusion, we evaluate UAD for paranasal anomaly detection. Previous methods [8][9][10] have used supervised learning methods and as a result are constrained to classify the anomalies that are included in the training distribution. Through our UAD approach, our models learn the healthy MS volume distribution X h thereby reducing the labelling effort of the clinicians.…”

“…There have been works of paranasal sinus anomaly diagnosis using supervised learning [10,8,11]. However, all these works have been used to classify at most one anomaly.…”

“…In a three year retrospective study, it was observed that malignant tumours and inverted papillomas were classified as nasal polyps with a misdiagnosis rate of 5.63% and 8.45% respectively. Therefore, computer aided diagnosis systems (CADx) [8][9][10] have been proposed with the idea of working in conjunction with the clinician to reduce the misdiagnosis rate of paranasal anomalies. However, all these works rely on supervised learning and consider at most one anomaly.…”

Desarrollo de una app móvil para la evaluación preoperatoria de la tomografía de senos paranasales: un paso hacia la inteligencia artificial

“…Algunas arquitecturas de aprendizaje profundo, como las redes neuronales profundas, las redes neuronales profundas convolucionales y los mecanismos de auto-atención, se han aplicado de forma exitosa en campos como la visión por computador y el procesamiento automático del lenguaje natural. En años recientes, estas tecnologías se han venido implementando en diferentes áreas de la medicina, y elsiendo el campo de la otorrinolaringología es una especialidad en la que se ha demostrado un creciente incremento de aplicaciones novedosas (2,4,5).…”