Abstract
Due to the huge mortality rate of lung cancer, there is a strong need for developing solutions that help with the early diagnosis and the definition of the most appropriate treatment. In the particular case of target therapy, effective genotyping of the tumor is fundamental since this treatment uses targeted drugs that can induce death in cancer cells. The biopsy is the traditional method to assess the genotype information but it is extremely invasive and painful. Medical imaging is a valuable alternative to biopsies, considering the potential to extract imaging features correlated with specific genomic alterations. Regarding the limitations of single model approaches for gene mutation status predictions, ensemble strategies might bring valuable benefits by combining the strengths and weaknesses of the aggregated methods. This preliminary work aims to provide further advances in the radiogenomics field by studying the use of ensemble methods to predict the Epidermal Growth Factor Receptor (EGFR) mutation status in lung cancer. The best result obtained for the proposed ensemble approach was an AUC of 0.706 (± 0.122). However, the ensemble did not outperform the single models with AUC values of 0.712 (± 0.119) for Logistic Regression, 0.711 (± 0.119) for Support Vector Machine and 0.712 (± 0.120) for Elastic Net. The high correlation found on the decisions of each single model might be a plausible explanation for this behavior, which caused the ensemble to misclassify the same examples as the single models.

Abstract
Glaucoma is a silent disease that shows symptoms when severe, leading to partial vision loss or irreversible blindness. Early screening permits treating patients in time. For glaucoma screening, retinal images are very important since they enable the observation of initial glaucoma lesions, which typically begins with the cupping formation in the optic disc (OD). In clinical settings, practical indicators such as Cup-to-Disc Ratio (CDR) are frequently used to evaluate the presence and stage of glaucoma. The ratio between the cup and the optic disc can be measured using the vertical or horizontal diameter, or the area of the two. Mass screening programs are limited by the high costs of specialised teams and equipment. Current deep learning (DL) methods can assist the glaucoma mass screening, lower the cost and allow it to be extended to larger populations. With DL methods in the OD and optic cup (OC) segmentation, is possible to evaluate the presence of glaucoma in the patient more quickly based on cupping formation in the OD, using CDR. In this work, is assessed the contribution of Multi-Class and Single-Class segmentation methods for glaucoma screening using the 3 types of CDR. U-Net architecture is trained using transfer learning models (Inception V3 and Inception ResNet V2) to segment the OD and OC and then evaluate glaucoma prediction based on different types of CDRs indicators. The models were trained and evaluated on main public known databases (REFUGE, RIM-ONE r3 and DRISHTI-GS). The segmentation of both OD and OC reach Dice over 0.8 and IoU above 0.7. The CDRs were computed to glaucoma assessment where was reach sensitivity above 0.8, specificity of 0.7, F1-Score around 0.7 and AUC above 0.85. Finally, conclusions of segmentation methods showing adequate performance to be used in practical glaucoma screening.

Abstract
Glaucoma is a silent disease characterized by progressive degeneration of retinal ganglion cells and, when not detected or treated early, can lead to blindness. Computer systems have demonstrated their efficiency in the medical decision-making process and Artificial Intelligence (AI) techniques have helped advances in ophthalmology, allowing for faster and more effective detection of glaucoma. Machine learning is a very promising subfield of AI that supports research in understanding the development, progression and treatment of glaucoma, identifying new risk factors and assessing the importance of existing ones. This study aims to test and analyze the results of different models of supervised machine learning in the detection and classification of ophthalmic diseases (Glaucoma, high myopia and low myopia) based on data from Corvis ST. The most important characteristics were selected based on a variance greater than 0.02. In terms of accuracy, the models that obtained the best results were Random Forrest 0.73, Stochastic Gradient Descent (SGD) 0.75, Gradient Boosting Classifier (GBC) 0.76 and K-Nearest Neighbors 0.71. The GBC model achieved the best results in accuracy, AUC, Recall and F1Score 76.00, 52.5, 78.00, 70.2 respectively.

Abstract
Capsule endoscopy made it possible to observe the inner lumen of the small bowel, but with the cost of a longer duration to process its resulting videos. Therefore, the scientific community has developed several machine learning strategies to help in detecting abnormalities in these videos. The published algorithms are typically trained and evaluated on small sets of images, ultimately not proving to be efficient when applied to full videos. In this experiment, we explored the problem of abnormality classification within an unbalanced dataset of images extracted from video capsule endoscopies, based on a vector feature extracted from the deepest layer of pre-trained Convolution Neural Networks to evaluate the impact of transfer learning with a small number of samples. The results showed that there is a reliable model on the classification task using small portions of data from video capsule endoscopies.

Abstract
The success achieved by using SAR data in the study of the Earth led to a firm commitment from space agencies to develop more and better space-borne SAR sensors. This involvement of the space agencies makes us believe that it is possible to increase the potential of SAR interferometry (InSAR) to near real-time monitoring. Among this ever-increasing number of sensors, the ESA's Sentinel-1 (C-band) mission stands out and appears to be disruptive. This mission is acquiring vast volumes of data making current analyzing approaches inviable. This amount of data can no longer be analyzed and studied using classic methods raising the need to use and create new techniques. We believe that Machine Learning techniques can be the solution to overcome this issue since they allow to train Deep Learning models to automate human processes for a vast volume of data. In this paper, we use deep learning models to automatically find and locate deformation areas in InSAR interferograms without atmospheric correction. We train three state-of-the-art classification models for detection deformation areas, achieving an AUC of 0.864 for the best model (VGG19 for wrapped interferograms). Additionally, we use the same models as encoders to train U-net models, achieving a Dice score of 0.54 for InceptionV3. It is necessary more data to achieve better results in segmentation.

Abstract
With an increasing interest in the digitization effort of ancient manuscripts, ancient character recognition becomes one of the most important areas in the automated document image analysis. In this regard, we propose a Convolutional Neural Network (CNN)-based classifier to recognize the ancient Sundanese characters obtained from a digital collection of Southeast Asian palm leaf manuscripts. In this work, we utilize two different preprocessing techniques for the dataset. The first technique involves the use of geometric transformations, noise background addition, and brightness adjustment to augment the imbalanced samples to be fed into the classifier. The second technique makes use of the Otsu's threshold method to binarize the characters and only uses the usual geometric transformations for the data augmentation. The proposed network with different data augmentation processes is trained on the training set and tested on the testing set. Image binarization from the second technique can outperform the performance of the CNN-based classifier upon the first technique by achieving a testing accuracy of 97.74%. (C) 2021 The Authors. Published by Elsevier B.V.