Skin Lesion Segmentation with Improved Convolutional Neural Network.

Melanoma is the most serious form of skin cancer, with more than 123,000 new cases worldwide each year. Reliable automatic melanoma screening system will be a great help for clinicians to detect malignant skin lesions as soon as possible. To address this problem, a computer-assisted approach to detecting melanoma using convolutional neural networks will enable effective and faster treatment of the disease. Many attempts have been made to use convolutional neural networks to solve this problem, but so far the performance in this aspect is not good. In this project, Xception and VGG-16 models will be fine-tuned on the IEEE International Symposium on Biomedical Imaging (ISBI) Official Skin Data Set 2016 and combined into an integrated framework to predict whether skin disease images are benign or malignant. Our experimental results show that the ensemble model can achieve better classification results and the proposed method is competitive in this field.

The IEEE International Symposium on Biomedical Imaging (ISBI) is a scientific conference dedicated to mathematical, algorithmic, and computational aspects of biological and biomedical imaging, across all scales of observation. It fosters knowledge transfer among different imaging communities and contributes to an integrative approach to biomedical imaging. ISBI is a joint initiative from the IEEE Signal Processing Society (SPS) and the IEEE Engineering in Medicine and Biology Society (EMBS). The 2018 meeting will include tutorials, and a scientific program composed of plenary talks, invited special sessions, challenges, as well as oral and poster presentations of peer-reviewed papers. High-quality papers are requested containing original contributions to the topics of interest including image formation and reconstruction, computational and statistical image processing and analysis, dynamic imaging, visualization, image quality assessment, and physical, biological, and statistical modeling. Accepted 4-page regular papers will be published in the symposium proceedings published by IEEE and included in IEEE Xplore. To encourage attendance by a broader audience of imaging scientists and offer additional presentation opportunities, ISBI 2018 will continue to have a second track featuring posters selected from 1-page abstract submissions without subsequent archival publication.

Meningioma is the commonest primary brain tumour. Volumetric post-contrast magnetic resonance imaging (MRI) is recognised as gold standard for delineation of meningioma volume but is hindered by manual processing times. We aimed to investigate the utility of a model-based variational approach in segmenting meningioma. A database of patients with a meningioma (2007-2015) was queried for patients with a contrast-enhanced volumetric MRI, who had consented to a research tissue biobank. Manual segmentation by a neuroradiologist was performed and results were compared to the mathematical model, using a battery of tests including the Sørensen-Dice coefficient (DICE) and JACCARD index. A publicly available meningioma dataset (708 segmented T1 contrast-enhanced slices) was also used to test the reliability of the model. 49 meningioma cases were included. The most common meningioma location was convexity (n = 15, 30.6%). The mathematical model segmented all but one incidental meningioma, which failed due to the lack of contrast uptake. The median meningioma volume by manual segmentation was 19.0cm3 (IQR 4.9-31.2). The median meningioma volume using the mathematical model was 16.9cm3 (IQR 4.6-28.34). The mean DICE score was 0.90 (SD = 0.04). The mean JACCARD index was 0.82 (SD = 0.07). For the publicly available dataset, the mean DICE and JACCARD scores were 0.90 (SD = 0.06) and 0.82 (SD = 0.10), respectively. Segmentation of meningioma volume using the proposed mathematical model was possible with accurate results. Application of this model on contrast-enhanced volumetric imaging may help reduce work burden on neuroradiologists with the increasing number in meningioma diagnoses.

Biological diversity is decreasing at a rate of 100-1000 times pre-human rates [1] [2], and tropical rainforests are among the most vulnerable ecosystems. To avoid species extinction, we need to understand factors influencing the occurrence of species. Fast, reliable computer-assisted tools can help to describe the habitat and thus to understand species habitat associations. This understanding is of utmost importance for more targeted species conservation efforts. Due to logistical challenges and time-consuming manual processing of field data, months up to years are often needed to progress from data collection to data interpretation. Deep learning can be used to significantly shorten the time while keeping a similar level of accuracy. Here, we propose Habitat-Net: a novel Convolutional Neural Network (CNN) based method to segment habitat images of rainforests. Habitat-Net takes color images as input and after multiple layers of convolution and deconvolution produces a binary segmentation of an image. The primary contribution of Habitat-Net is the translation of medical imaging knowledge (inspired by U-Net [3]) to ecological problems. The entire Habitat-Net pipeline works automatically without any user interaction. Our only assumption is the availability of annotated images, from which Habitat-Net learns the most distinguishing features automatically. In our experiments, we use two habitat datasets: (1) canopy and (2) understory vegetation. We train the model with 800 canopy images and 700 understory images separately. Our testing dataset has 150 canopy and 170 understory images. We use the Dice coefficient and Jaccard Index to quantify the overlap between ground-truthed segmentation images and those obtained by Habitat-Net model. This results in a mean Dice Score (mean Jaccard Index) for the segmentation of canopy and understory images of 0.89 (0.81) and 0.79 (0.69), respectively. Compared to manual segmentation, Habitat-Net prediction is approximately 3K – 150K times faster. For a typical canopy dataset of 335 images, Habitat-Net reduces total processing time to 5 seconds (15 milliseconds/ image) from 4 hours (45 seconds/ image). In this study, we show that it is possible to speed up the data pipeline using deep learning in the ecological domain. In the future, we plan to create a freely available mobile app based on Habitat-Net technology to characterize the habitat directly and automated in the field. In combination with ecological models our tools will help to understand the ecology of some poorly known, but often highly threatened, species and thus contribute to more timely conservation interventions.
\nREFERENCES: 
\n1. Sachs et al. "Biodiversity conservation and the millennium development goals." Science 325.5947 (2009): 1502-1503.
\n2. Chapin Iii, F. Stuart, et al. "Consequences of changing biodiversity." Nature 405.6783 (2000): 234.
\n3. Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015.

Tracking subcellular structures as well as viral structures displayed as 'particles' in fluorescence microscopy images yields quantitative information on the underlying dynamical processes. We have developed an approach for tracking multiple fluorescent particles based on probabilistic data association. The approach combines a localization scheme that uses a bottom-up strategy based on the spot-enhancing filter as well as a top-down strategy based on an ellipsoidal sampling scheme that uses the Gaussian probability distributions computed by a Kalman filter. The localization scheme yields multiple measurements that are incorporated into the Kalman filter via a combined innovation, where the association probabilities are interpreted as weights calculated using an image likelihood. To track objects in close proximity, we compute the support of each image position relative to the neighboring objects of a tracked object and use this support to recalculate the weights. To cope with multiple motion models, we integrated the interacting multiple model algorithm. The approach has been successfully applied to synthetic 2-D and 3-D images as well as to real 2-D and 3-D microscopy images, and the performance has been quantified. In addition, the approach was successfully applied to the 2-D and 3-D image data of the recent Particle Tracking Challenge at the IEEE International Symposium on Biomedical Imaging (ISBI) 2012.

Land cover is an important variable of the terrestrial ecosystem that provides information for natural resources management, urban sprawl detection, and environment research. To classify land cover with high-spatial-resolution multispectral remote sensing imagery is a difficult problem due to heterogeneous spectral values of the same object on the ground. Fully convolutional networks (FCNs) are a state-of-the-art method that has been increasingly used in image segmentation and classification. However, a systematic quantitative comparison of FCNs on high-spatial-multispectral remote imagery was not yet performed. In this paper, we adopted the three FCNs (FCN-8s, Segnet, and Unet) for Gaofen-2 (GF2) satellite imagery classification. Two scenes of GF2 with a total of 3329 polygon samples were used in the study area and a systematic quantitative comparison of FCNs was conducted with red, green, blue (RGB) and RGB+near infrared (NIR) inputs for GF2 satellite imagery. The results showed that: (1) The FCN methods perform well in land cover classification with GF2 imagery, and yet, different FCNs architectures exhibited different results in mapping accuracy. The FCN-8s model performed best among the Segnet and Unet architectures due to the multiscale feature channels in the upsampling stage. Averaged across the models, the overall accuracy (OA) and Kappa coefficient (Kappa) were 5% and 0.06 higher, respectively, in FCN-8s when compared with the other two models. (2) High-spatial-resolution remote sensing imagery with RGB+NIR bands performed better than RGB input at mapping land cover, and yet the advantage was limited; the OA and Kappa only increased an average of 0.4% and 0.01 in the RGB+NIR bands. (3) The GF2 imagery provided an encouraging result in estimating land cover based on the FCN-8s method, which can be exploited for large-scale land cover mapping in the future.

Hyper-fusion network for semi-automatic segmentation of skin lesions.

Skin lesion segmentation plays an important role in automatic skin cancer diagnosis. One of the most dangerous types of this cancer is melanoma which requires an early and accurate detection. However, automatic melanoma segmentation on dermoscopic images is a challenging task since images are corrupted by noise like hairs, air bubbles, blood vessel… and with fuzzy boundaries. This paper presents a framework based on deep fully convolutional neural networks to automatically segment skin lesions in dermoscopic images. Particularly, the paper proposes a fully convolutional network (FCN) framework that is based on modification of the SegNet architecture. In particular, we propose to reduce the downsampling and upsampling layers in the original SegNet model, that reduces total learned parameters compared to the original SegNet model. The proposed approach is applied to segment images from ISIC 2017 dataset. Experimental results show the desired performances of the proposed approach in terms of metrics of Dice coefficient and Jaccard indexes.

Cardiac Magnetic Resonance (CMR) imaging is commonly used to assess cardiac structure and function. One disadvantage of CMR is that postprocessing of exams is tedious. Without automation, precise assessment of cardiac function via CMR typically requires an annotator to spend tens of minutes per case manually contouring ventricular structures. Automatic contouring can lower the required time per patient by generating contour suggestions that can be lightly modified by the annotator. Fully convolutional networks (FCNs), a variant of convolutional neural networks, have been used to rapidly advance the state-of-the-art in automated segmentation, which makes FCNs a natural choice for ventricular segmentation. However, FCNs are limited by their computational cost, which increases the monetary cost and degrades the user experience of production systems. To combat this shortcoming, we have developed the FastVentricle architecture, an FCN architecture for ventricular segmentation based on the recently developed ENet architecture. FastVentricle is 4\(\times \) faster and runs with 6\(\times \) less memory than the previous state-of-the-art ventricular segmentation architecture while still maintaining excellent clinical accuracy.

AFLN-DGCL: Adaptive Feature Learning Network with Difficulty-Guided Curriculum Learning for skin lesion segmentation

The semantic segmentation of skin lesions is an important and common initial task in the computer aided diagnosis of dermoscopic images. Although deep learning-based approaches have considerably improved the segmentation accuracy, there is still room for improvement by addressing the major challenges, such as variations in lesion shape, size, color and varying levels of contrast. In this work, we propose the first deep semantic segmentation framework for dermoscopic images which incorporates, along with the original RGB images, information extracted using the physics of skin illumination and imaging. In particular, we incorporate information from specific color bands, illumination invariant grayscale images, and shading-attenuated images. We evaluate our method on three datasets: the ISBI ISIC 2017 Skin Lesion Segmentation Challenge dataset, the DermoFit Image Library, and the PH2 dataset and observe improvements of 12.02%, 4.30%, and 8.86% respectively in the mean Jaccard index over a baseline model trained only with RGB images.

Farmland is the cornerstone of agriculture and is important for food security and social production. Farmland assessment is essential but traditional methods are usually expensive and slow. Deep learning methods have been developed and widely applied recently in image recognition, semantic understanding, and many other application domains. In this research, we used fully convolutional networks (FCN) as the deep learning model to evaluate farmland grades. Normalized difference vegetation index (NDVI) derived from Landsat images was used as the input data, and the China National Cultivated Land Grade Database within Jiangsu Province was used to train the model on cloud computing. We also applied an image segmentation method to improve the original results from the FCN and compared the results with classical machine learning (ML) methods. Our research found that the FCN can predict farmland grades with an overall F1 score (the harmonic mean of precision and recall) of 0.719 and F1 score of 0.909, 0.590, 0.740, 0.642, and 0.023 for non-farmland, level I, II, III, and IV farmland, respectively. Combining the FCN and image segmentation method can further improve prediction accuracy with results of fewer noise pixels and more realistic edges. Compared with conventional ML, at least in farmland evaluation, FCN provides better results with higher precision, recall, and F1 score. Our research indicates that by using remote sensing NDVI data, the deep learning method can provide acceptable farmland assessment without fieldwork and can be used as a novel supplement to traditional methods. The method used in this research will save a lot of time and cost compared with traditional means.

Skin lesion classification using CNNs with grouping of multi-scale attention and class-specific loss weighting

In technologically underdeveloped areas, water pollution threatens the living environment of local residents, so remote sensing monitoring of the features around reservoirs is necessary. Fully convolutional networks (FCNs) offer great potential for extracting high-resolution features due to their unlimited input image size and higher accuracy compared to convolutional neural networks. Therefore, a proposal to classify WorldView-2 images is implemented with a sixty-eight thousand iterations of fine-tuning and fully trained combined training method based on a fully convolutional network (SEFCN). The chosen images depict the urban area of Yingde, which is located to the northeast of the Feilaixia Reservoir, Qingyuan, Guangdong Province, China. The SEFCN combines an FCN-32s and FCN-16s to better integrate the deep features and shallow features and effectively improve the classification accuracy. Additionally, the loss value fully converges with enough iterations, and the overfitting caused by the superposition of two full trainings is avoided. The SEFCN model achieves the highest accuracy among all compared classification models and obtains the best classified results on the WorldView-2 images, as demonstrated by attaining the highest F1 scores in each category. After the classified images are optimized using a conditional random field, the current status of the study area is analyzed, and several suggestions for land use in the urban area of Yingde are made. The experiments still have deficiencies in the application of high-resolution remote sensing image classification with different sensors, and the classification results can be optimized and improved in other aspects.

Glaucoma, the leading cause of irreversible blindness, must be diagnosed early and thus treated in time. However, it has no noticeable symptoms in its early stages and may not be detected easily. This paper aims to compare two well-known convolutional neural network (CNN) structures, namely Fully Convolutional Networks (FCNs) and U-Net for the segmentation of the optic disc (OD) and optic cup (OC) from retinal fundus images which play an important role in glaucoma diagnosis. The performance of both models is assessed using qualitative parameters such as the Dice coefficient, Jaccard index, and cup-to-disc ratio (CDR) error. In our experiment, the U-Net model yields more accurate segmentation results with 0.9601 average pixel accuracy and 0.9255 dice score for OD segmentation, outperforming the FCNs model with 0.9560 average pixel accuracy and 0.9132 dice score for OD segmentation. However, FCNs have a shorter inference time of 0. 0043 seconds against U-net’s 0. 0062 seconds making FCNs more suitable for real-time applications. The restrictions related to this study include biases from using only one dataset acquired from particular imaging devices, dependency on mask-based cropping techniques, and comparison being restricted to two fundamental architectures. This work presents the contribution of the deep learning models in improving glaucoma screening and therefore helping in avoiding blindness.