The three dimensional dosimetry imaging for automated eye cancer classification using transfer learning model

Computer vision methods utilizing transfer learning can offer a promising approach for automated eye cancer diagnosis with three-dimensional dosimetry imaging. The latest advances in AI have encouraged researchers to use deep learning models for automated cancer classification. This paper presents a novel approach using three-dimensional dosimetry imaging for automated ocular cancer classification using transfer learning. The approach uses a combination of dataset augmentation methods. The approach is adapted from the popular Inception V3 CNN architecture. Two different feature extraction approaches, namely mean and maximum pooling, are compared while extracting features from the output of the CNN layers. The extracted features are fed to the classifier which uses the SVM with a linear kernel as its decision system. The effectiveness of this method is evaluated on a set of 20 healthy ocular images and 26 suspicious ocular images. The overall classification accuracy achieved with this approach is 94.85%. The results obtained demonstrate that this approach has a promising potential for automated ocular cancer classification with an appropriate transfer learning model.