Benign vs malignant tumors classification from tumor outlines in mammography scans using artificial intelligence techniques.

Brain tumour detection and classification require trained radiologists for efficient diagnosis. The proposed work aims to build a Computer Aided Diagnosis (CAD) tool to automate brain tumour detection using Machine Learning (ML) and Deep Learning (DL) techniques. Magnetic Resonance Image (MRI) collected from the publicly available Kaggle dataset is used for brain tumour detection and classification. Deep features extracted from the global pooling layer of Pretrained Resnet18 network are classified using 3 different ML Classifiers, such as Support vector Machine (SVM), K-Nearest Neighbour (KNN), and Decision Tree (DT). The above classifiers are further hyperparameter optimised using Bayesian Algorithm (BA) to enhance the performance. Fusion of features extracted from shallow and deep layers of the pretrained Resnet18 network followed by BA-optimised ML classifiers is further used to enhance the detection and classification performance. The confusion matrix derived from the classifier model is used to evaluate the system's performance. Evaluation metrics, such as accuracy, sensitivity, specificity, precision, F1 score, Balance Classification Rate (BCR), Mathews Correlation Coefficient (MCC) and Kappa Coefficient (Kp), are calculated. Maximum accuracy, sensitivity, specificity, precision, F1 score, BCR, MCC, and Kp of 99.11 %, 98.99 %, 99.22 %, 99.09 %, 99.09 %, 99.10 %, 98.21 %, 98.21 %, respectively, were obtained for detection using fusion of shallow and deep features of Resnet18 pretrained network classified by BA optimized SVM classifier. Feature fusion performs better for classification task with accuracy, sensitivity, specificity, precision, F1 score, BCR, MCC and Kp of 97.31 %, 97.30 %, 98.65 %, 97.37 %, 97.34 %, 97.97%, 95.99 %, 93.95 %, respectively. The proposed brain tumour detection and classification framework using deep feature extraction from Resnet 18 pretrained network in conjunction with feature fusion and optimised ML classifiers can improve the system performance. Henceforth, the proposed work can be used as an assistive tool to aid the radiologist in automated brain tumour analysis and treatment.

This paper addresses automated glaucoma detection system using pre-trained convolutional neural networks (CNNs). CNNs, a class of deep neural networks (DNNs), extract features of high-level abstractions from the fundus images, thereby eliminating the need for hand-crafted features which are prone to inaccuracies in segmenting landmark regions and require excessive involvement of experts for annotating these landmarks. This work investigates the applicability of pre-trained CNNs for glaucoma diagnosis, which is preferred when the dataset size is small. Further, pre-trained networks have the advantage of the quick model building. The proposed system has been validated on the High-Resolution (HRF), which is a publicly available benchmark database. Results demonstrate that among other pre-trained CNNs, VGG16 network is more suitable for glaucoma diagnosis.

In this paper, we investigate the performance of two deep learning paradigms for the audio-based tasks of acoustic scene, environmental sound and domestic activity classification. In particular, a convolutional recurrent neural network (CRNN) and pre-trained convolutional neural networks (CNNs) are utilised. The CRNN is directly trained on Mel-spectrograms of the audio samples. For the pre-trained CNNs, the activations of one of the top layers of various architectures are extracted as feature vectors and used for training a linear support vector machine (SVM).Moreover, the predictions of the two models—the class probabilities predicted by the CRNN and the decision function of the SVM—are combined in a decision-level fusion to achieve the final prediction. For the pre-trained CNN networks we use as feature extractors, we further evaluate the effects of a range of configuration options, including the choice of the pre-training corpus. The system is evaluated on the acoustic scene classification task of the IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE 2017) workshop, ESC-50 and the multi-channel acoustic recordings from DCASE 2018, task 5. We have refrained from additional data augmentation as our primary goal is to analyse the general performance of the proposed system on different datasets. We show that using our system, it is possible to achieve competitive performance on all datasets and demonstrate the complementarity of CRNNs and ImageNet pre-trained CNNs for acoustic classification tasks. We further find that in some cases, CNNs pre-trained on ImageNet can serve as more powerful feature extractors than AudioSet models. Finally, ImageNet pre-training is complimentary to more domain-specific knowledge, either in the form of the convolutional recurrent neural network (CRNN) trained directly on the target data or the AudioSet pre-trained models. In this regard, our findings indicate possible benefits of applying cross-modal pre-training of large CNNs to acoustic analysis tasks.

Utilizing pretrained convolutional neural networks for crack detection and geometric feature recognition in concrete surface images

Diabetic retinopathy is the leading cause of blindness, engaging people in different ages. Early detection of the disease, although significantly important to control and cure it, is usually being overlooked due to the need for experienced examination. To this end, automatic diabetic retinopathy diagnostic methods are proposed to facilitate the examination process and act as the physician's helper. In this paper, automatic diagnosis of diabetic retinopathy using pre-trained convolutional neural networks is studied. Pre-trained networks are chosen to avoid the time-and resource-consuming training algorithms for designing a convolutional neural network from scratch. Each neural network is fine-tuned with the pre-processed dataset, and the fine-tuning parameters as well as the pre-trained neural networks are compared together. The result of this paper, introduces a fast approach to fine-tune pre-trained networks, by studying different tuning parameters and their effect on the overall system performance due to the specific application of diabetic retinopathy screening.

This paper proposes new features extracted from images derived from flow, for first-person activity recognition. Features from convolutional neural network (CNN), which is designed for 2D images, attract attention from computer vision researchers due to its powerful discrimination capability, and recently a convolutional neural network for videos, called C3D (Convolutional 3D), was proposed. Generally CNN / C3D features are extracted directly from original images / videos with pre-trained convolutional neural network, since the network was trained with images / videos. In this paper, on the other hand, we propose the use of images derived from flow (we call this image as optical flow image) as input images into the pre-trained neural network, based on the following reasons; (i) flow images give dynamic information which is useful for activity recognition, compared with original images, which give only static information, and (ii) the pre-trained network has chance to extract features with reasonable discrimination capability, since the network was trained with huge amount of images from big categories. We carry out experiments with a dataset named DogCentric Activity Dataset, and we show the effectiveness of the extracted features.

Categorization of scene images is considered as a challenging prospect due to the fact that different classes of scene images often share similar image statistics. This chapter presents a transfer learning based approach for scene classification. A pre-trained Convolutional Neural Network (CNN) is used as a feature extractor for the images. The pre-trained network along with classifiers such as Support Vector Machines (SVM) or Multi Layer Perceptron (MLP) are used to classify the images. Also, the effect of single plane images such as, RGB2Gray, SVD Decolorized and Modified SVD decolorized images are analysed based on classification accuracy, class-wise precision, recall, F1-score and equal error rate (EER). The classification experiment for SVM was also done using a dimensionality reduction technique known as principal component analysis (PCA) on the feature vector. By comparing the results of models trained on RGB images with those grayscale images, the difference in the results is very small. These grayscale images were capable of retaining the required shape and texture information from the original RGB images and were also sufficient to categorize the classes of the given scene images.

Age estimation based on face images plays an important role in a wide range of scenarios, including security and defense applications, border control, human-machine interaction in ambient intelligence applications, and recognition based on soft biometric information. Recent methods based on deep learning have shown promising performance in this field. Most of these methods use deep networks specifically designed and trained to cope with this problem. There are also some studies that focus on applying deep networks pre-trained for face recognition, which perform a fine-tuning to achieve accurate results. Differently, in this paper, we propose a preliminary study on increasing the performance of pre-trained deep networks by applying postprocessing strategies. The main advantage with respect to fine-tuning strategies consists of the simplicity and low computational cost of the post-processing step. To the best of our knowledge, this paper is the first study on age estimation that proposes the use of post-processing strategies for features extracted using pre-trained deep networks. Our method exploits a set of pre-trained Convolutional Neural Networks (CNNs) to extract features from the input face image. The method then performs a feature level fusion, reduces the dimensionality of the feature space, and estimates the age of the individual by using a Feed-Forward Neural Network (FFNN). We evaluated the performance of our method on a public dataset (Adience Benchmark of Unfiltered Faces for Gender and Age Classification) and on a dataset of nonideal samples affected by controlled rotations, which we collected in our laboratory. Our age estimation method obtained better or comparable results with respect to state-of-the-art techniques and achieved satisfactory performance in non-ideal conditions. Results also showed that CNNs trained on general datasets can obtain satisfactory accuracy for different types of validation images, also without applying fine-tuning methods.

Any disease or disorder in the human body if recognized at an early stage, effective treatment can be given. Emphysema is a type of CPOD (Chronic obstructive pulmonary disease) and it is due to malfunctioning of lungs. Breathing becomes difficult with this disease. It occurs due to the stretching and damaging of air sacs in the lungs. In this work, an ensemble approach is proposed using the features extracted from the last global average pooling layer of the pre trained convolutional neural networks which are trained on `Imagenet' data set. These features are given to an Error Correcting Output Code classifier with the base classifier being Support Vector Machine. Three best pre trained networks are selected based on the average classification accuracy. An ensemble of the three classifiers is considered. Based on the majority voting, weighed average probability and highest probability strategy, the test images labels are identified. Hold out validation (80% training and 20% testing) is used to assess the performance of the proposed algorithm. A popular database of computed tomography emphysema images is chosen to validate our proposal. A peak classification accuracy of 100% and an average classification accuracy of 95.88% is obtained with a combination of Resnet18, Shufflenet, and Resnet 101 pre trained networks with the averaged probability as a choice in the prediction of labels of test images. Compared to the state of the art approaches for classifying emphysema, the proposed method is superior in terms of classification accuracy.

Diabetic retinopathy is a serious complication of diabetes, and if not controlled, may cause blindness. Automated screening of diabetic retinopathy helps physicians to diagnose and control the disease in early stages. In this paper, two case studies are proposed, each on a different dataset. Firstly, automatic screening of diabetic retinopathy utilising pre-trained convolutional neural networks was employed on the Kaggle dataset. The reason for using pre-trained networks is to save time and resources during training compared to fully training a convolutional neural network. The proposed networks were fine-tuned for the pre-processed dataset, and the selectable parameters of the fine-tuning approach were optimised. At the end, the performance of the fine-tuned network was evaluated using a clinical dataset comprising 101 images. The clinical dataset is completely independent from the fine-tuning dataset and is taken by a different device with different image quality and size.

Due to drastic changes in the field of technology and computing power for the last decade, it has become very easy to implement the convolutional neural networks for the classification and detection of objects from the large volume of images. Nowadays, the various deep networks with hundreds of layers are developed and implemented by the researchers for the classification of images and object detection inside the images. The Faster region-based convolutional neural network (R-CNN) is a widely used state-of-the-art approach that belongs to R-CNN techniques that were first time developed and used in 2015. Different R-CNN object detection approaches are developed and implemented by the researchers. Three approaches are developed and implemented on different platforms, and these approaches are R-CNN, fast R-CNN, and faster R-CNN. The efficiency and accuracy of the approaches are tested for various object detections inside the different images. Algorithms based on region proposals are used in R-CNN approaches to generate the bounding boxes or the actual location of the objects inside the images. The ground labels are generated through image labeling approaches. These ground truth labels are stored in a file. The features are extracted by pre-trained deep networks or the convolutional neural networks using the ground truth labeled images. The classification layer of the convolutional neural networks predicts the class of the object to which it belongs. The regression layer is used to create the relevant coordinates of the bounding boxes accurately. In this research paper, the faster R-CNN approach with retrained deep networks is used for the detection of pituitary tumor. The tumor detection performance of the detectors trained with three pre-trained deep networks is compared in the proposed approach of tumor detection. Three pre-trained deep networks such as Googlenet, Resnet18, and Resnet50 are used to train the tumor detector with ground truth labeled images.

The process of classification of fruits and their manual gradation, is an age-old phenomena taking place since centuries, and has been there in almost all the fruit markets. But with the advent of handy digital cameras and improvement in the image capturing and processing systems, the age-old manual classification methods have been fast replaced by the automatic classification techniques supported by computer vision based techniques. The thermal imaging is one of the popular non-destructive methodology applied to classify fruits. In the automatic classification of images machine learning techniques are being replaced by deep learning techniques because of their higher classification accuracy rates. In the present research work, a thermal image dataset has been created for eleven different varieties of fruits and classified using pre-trained convolutional neural networks (CNN). In our work, instead of building and training a CNN from start, we have utilized a pre-built and pre-trained network via transfer learning. Parameters like sensitivity, specificity, F1-score, precision and accuracy of CNN for transfer learning of fruit classification have been tested. The results reveal that our classification system has Top-1 and Top-5 accuracy as 96.54% and 100% respectively with the training time of 9.21 minutes using SqueezeNet model.

Objective. Prediction of the response to neoadjuvant chemotherapy (NAC) in breast cancer is important for patient outcomes. In this work, we propose a deep learning based approach to NAC response prediction in ultrasound (US) imaging. Approach. We develop recurrent neural networks that can process serial US imaging data to predict chemotherapy outcomes. We present models that can process either raw radio-frequency (RF) US data or regular US images. The proposed approach is evaluated based on 204 sequences of US data from 51 breast cancers. Each sequence included US data collected before the chemotherapy and after each subsequent dose, up to the 4th course. We investigate three pre-trained convolutional neural networks (CNNs) as back-bone feature extractors for the recurrent network. The CNNs were pre-trained using raw US RF data, US b-mode images and RGB images from the ImageNet dataset. The first two networks were developed using US data collected from malignant and benign breast masses. Main results. For the pre-treatment data, the better performing network, with back-bone CNN pre-trained on US images, achieved area under the receiver operating curve (AUC) of 0.81 (±0.04). Performance of the recurrent networks improved with each course of the chemotherapy. For the 4th course, the better performing model, based on the CNN pre-trained with RGB images, achieved AUC value of 0.93 (±0.03). Statistical analysis based on the DeLong test presented that there were no significant differences in AUC values between the pre-trained networks at each stage of the chemotherapy (p-values > 0.05). Significance. Our study demonstrates the feasibility of using recurrent neural networks for the NAC response prediction in breast cancer US.

Texture classification plays a vital role in the emerging research field of image classification. This paper approaches the texture classification problem using significant features extracted from pre-trained Convolutional Neural Network (CNN) like Alexnet, VGG16, Resnet18, Googlenet, MobilenetV2, and Darknet19. These features are classified by machine learning classifiers such as Support Vector Machine (SVM), Ensemble, K Nearest Neighbour (KNN), Naïve Bayes (NB), Decision Tree (DT), and Discriminant Analysis (DA). The performance of the work is evaluated with the texture databases namely KTH-TIPS, FMD, UMD-HR, and DTD. Among these CNN features derived from VGG16 classify by SVM provides better classification accuracy rather than using VGG16 with a softmax classifier.

With the help of machine learning, many of the problems that have plagued mammography in the past have been solved. Effective prediction models need many normal and tumor samples. For medical applications such as breast cancer diagnosis framework, it is difficult to gather labeled training data and construct effective learning frameworks. Transfer learning is an emerging strategy that has recently been used to tackle the scarcity of medical data by transferring pre-trained convolutional network knowledge into the medical domain. Despite the well reputation of the transfer learning based on the pre-trained Convolutional Neural Networks (CNN) for medical imaging, several hurdles still exist to achieve a prominent breast cancer classification performance. In this paper, we attempt to solve the Feature Dimensionality Curse (FDC) problem of the deep features that are derived from the transfer learning pre-trained CNNs. Such a problem is raised due to the high space dimensionality of the extracted deep features with respect to the small size of the available medical data samples. Therefore, a novel deep learning cascaded feature selection framework is proposed based on the pre-trained deep convolutional networks as well as the univariate-based paradigm. Deep learning models of AlexNet, VGG, and GoogleNet are randomly selected and used to extract the shallow and deep features from the INbreast mammograms, whereas the univariate strategy helps to overcome the dimensionality curse and multicollinearity issues for the extracted features. The optimized key features via the univariate approach are statistically significant (p-value ≤ 0.05) and have good capability to efficiently train the classification models. Using such optimal features, the proposed framework could achieve a promising evaluation performance in terms of 98.50% accuracy, 98.06% sensitivity, 98.99% specificity, and 98.98% precision. Such performance seems to be beneficial to develop a practical and reliable computer-aided diagnosis (CAD) framework for breast cancer classification.