Supervised Saliency Map Driven Segmentation of Lesions in Dermoscopic Images

Boosting the performance of pretrained CNN architecture on dermoscopic pigmented skin lesion classification

BACKGROUND

Pigmented skin lesions (PSLs) pose medical and esthetic challenges for those affected. PSLs can cause skin cancers, particularly melanoma, which can be life-threatening. Detecting and treating melanoma early can reduce mortality rates. Dermoscopic imaging offers a noninvasive and cost-effective technique for examining PSLs. However, the lack of standardized colors, image capture settings, and artifacts makes accurate analysis challenging. Computer-aided diagnosis (CAD) using deep learning models, such as convolutional neural networks (CNNs), has shown promise by automatically extracting features from medical images. Nevertheless, enhancing the CNN models' performance remains challenging, notably concerning sensitivity.

MATERIALS AND METHODS

In this study, we aim to enhance the classification performance of selected pretrained CNNs. We use the 2019 ISIC dataset, which presents eight disease classes. To achieve this goal, two methods are applied: resolution of the dataset imbalance challenge through augmentation and optimization of the training hyperparameters via Bayesian tuning.

RESULTS

The performance improvement was observed for all tested pretrained CNNs. The Inception-V3 model achieved the best performance compared to similar results, with an accuracy of 96.40% and an AUC of 0.98.

CONCLUSION

According to the study, classification performance was significantly enhanced by augmentation and Bayesian hyperparameter tuning.

Deep Hypersphere Feature Regularization for Weakly Supervised RGB-D Salient Object Detection

We propose a weakly supervised approach for salient object detection from multi-modal RGB-D data. Our approach only relies on labels from scribbles, which are much easier to annotate, compared with dense labels used in conventional fully supervised setting. In contrast to existing methods that employ supervision signals on the output space, our design regularizes the intermediate latent space to enhance discrimination between salient and non-salient objects. We further introduce a contour detection branch to implicitly constrain the semantic boundaries and achieve precise edges of detected salient objects. To enhance the long-range dependencies among local features, we introduce a Cross-Padding Attention Block (CPAB). Extensive experiments on seven benchmark datasets demonstrate that our method not only outperforms existing weakly supervised methods, but is also on par with several fully-supervised state-of-the-art models. Code is available at https://github.com/leolyj/DHFR-SOD.

Dermatologist versus artificial intelligence confidence in dermoscopy diagnosis: Complementary information that may affect decision-making

In dermatology, deep learning may be applied for skin lesion classification. However, for a given input image, a neural network only outputs a label, obtained using the class probabilities, which do not model uncertainty. Our group developed a novel method to quantify uncertainty in stochastic neural networks. In this study, we aimed to train such network for skin lesion classification and evaluate its diagnostic performance and uncertainty, and compare the results to the assessments by a group of dermatologists. By passing duplicates of an image through such a stochastic neural network, we obtained distributions per class, rather than a single probability value. We interpreted the overlap between these distributions as the output uncertainty, where a high overlap indicated a high uncertainty, and vice versa. We had 29 dermatologists diagnose a series of skin lesions and rate their confidence. We compared these results to those of the network. The network achieved a sensitivity and specificity of 50% and 88%, comparable to the average dermatologist (respectively 68% and 73%). Higher confidence/less uncertainty was associated with better diagnostic performance both in the neural network and in dermatologists. We found no correlation between the uncertainty of the neural network and the confidence of dermatologists (R = -0.06, p = 0.77). Dermatologists should not blindly trust the output of a neural network, especially when its uncertainty is high. The addition of an uncertainty score may stimulate the human-computer interaction.

Generative adversarial networks based skin lesion segmentation

Skin cancer is a serious condition that requires accurate diagnosis and treatment. One way to assist clinicians in this task is using computer-aided diagnosis tools that automatically segment skin lesions from dermoscopic images. We propose a novel adversarial learning-based framework called Efficient-GAN (EGAN) that uses an unsupervised generative network to generate accurate lesion masks. It consists of a generator module with a top-down squeeze excitation-based compound scaled path, an asymmetric lateral connection-based bottom-up path, and a discriminator module that distinguishes between original and synthetic masks. A morphology-based smoothing loss is also implemented to encourage the network to create smooth semantic boundaries of lesions. The framework is evaluated on the International Skin Imaging Collaboration Lesion Dataset. It outperforms the current state-of-the-art skin lesion segmentation approaches with a Dice coefficient, Jaccard similarity, and accuracy of 90.1%, 83.6%, and 94.5%, respectively. We also design a lightweight segmentation framework called Mobile-GAN (MGAN) that achieves comparable performance as EGAN but with an order of magnitude lower number of training parameters, thus resulting in faster inference times for low compute resource settings.

Attention-Guided Network with Densely Connected Convolution for Skin Lesion Segmentation

The automatic segmentation of skin lesions is considered to be a key step in the diagnosis and treatment of skin lesions, which is essential to improve the survival rate of patients. However, due to the low contrast, the texture and boundary are difficult to distinguish, which makes the accurate segmentation of skin lesions challenging. To cope with these challenges, this paper proposes an attention-guided network with densely connected convolution for skin lesion segmentation, called CSAG and DCCNet. In the last step of the encoding path, the model uses densely connected convolution to replace the ordinary convolutional layer. A novel attention-oriented filter module called Channel Spatial Fast Attention-guided Filter (CSFAG for short) was designed and embedded in the skip connection of the CSAG and DCCNet. On the ISIC-2017 data set, a large number of ablation experiments have verified the superiority and robustness of the CSFAG module and Densely Connected Convolution. The segmentation performance of CSAG and DCCNet is compared with other latest algorithms, and very competitive results have been achieved in all indicators. The robustness and cross-data set performance of our method was tested on another publicly available data set PH2, further verifying the effectiveness of the model.

A Probabilistic-Based Deep Learning Model for Skin Lesion Segmentation

The analysis and detection of skin cancer diseases from skin lesion have always been tedious when done manually. The complex nature of skin lesion images is one of the key reasons for this. The skin lesion images contain noise and artifacts such as hairs, oil and bubbles, blood vessels, and skin lines. They also have variegated colors, low contrast, and irregular borders. Various computational approaches have been designed in the past for aiding in the detection and diagnosis of skin cancer diseases using skin lesion images. The existing techniques have been limited due to the interference of the aforementioned features of skin lesion. Recently, machine learning techniques, in particular the deep learning techniques have been used for the detection of skin cancer. However, they are still limited to the fuzzy and irregular borders of skin lesion images coupled with the low contrast that exists between the diseased lesion and healthy tissues. In this paper, we utilized a probabilistic model for the enhancement of a fully convolutional network-based deep learning system to analyze and segment skin lesion images. The probabilistic model employs an efficient mean-field approximate probabilistic inference approach with a fully connected conditional random field that utilizes a Gaussian kernel. The probabilistic model further performs a refinement of skin lesion borders. The whole framework is tested and evaluated on publicly available skin lesion image datasets of ISBI 2017 and PH2. The system achieved a better performance, having an accuracy of 98%.

Towards the automatic detection of skin lesion shape asymmetry, color variegation and diameter in dermoscopic images

Asymmetry, color variegation and diameter are considered strong indicators of malignant melanoma. The subjectivity inherent in the first two features and the fact that 10% of melanomas tend to be missed in the early diagnosis due to having a diameter less than 6mm, deem it necessary to develop an objective computer vision system to evaluate these criteria and aid in the early detection of melanoma which could eventually lead to a higher 5-year survival rate. This paper proposes an approach for evaluating the three criteria objectively, whereby we develop a measure to find asymmetry with the aid of a decision tree which we train on the extracted asymmetry measures and then use to predict the asymmetry of new skin lesion images. A range of colors that demonstrate the suspicious colors for the color variegation feature have been derived, and Feret's diameter has been utilized to find the diameter of the skin lesion. The decision tree is 80% accurate in determining the asymmetry of skin lesions, and the number of suspicious colors and diameter values are objectively identified.

New Auxiliary Function with Properties in Nonsmooth Global Optimization for Melanoma Skin Cancer Segmentation

In this paper, an algorithm is introduced to solve the global optimization problem for melanoma skin cancer segmentation. The algorithm is based on the smoothing of an auxiliary function that is constructed using a known local minimizer and smoothed by utilising Bezier curves. This function achieves all filled function properties. The proposed optimization method is applied to find the threshold values in melanoma skin cancer images. The proposed algorithm is implemented on PH2, ISBI2016 challenge, and ISBI 2017 challenge datasets for melanoma segmentation. The results show that the proposed algorithm exhibits high accuracy, sensitivity, and specificity compared with other methods.

Skin lesion segmentation using high-resolution convolutional neural network

BACKGROUND AND OBJECTIVE

Skin lesion segmentation is an important but challenging task in computer-aided diagnosis of dermoscopy images. Many segmentation methods based on convolutional neural networks often fail to extract accurate lesion boundaries because the spatial size of feature maps decreases as the maps are processed throughout the network layers. We propose skin lesion segmentation in dermoscopy images based on a convolutional neural network with an attention mechanism, which can preserve edge details.

METHODS

We devised a high-resolution feature block containing three branches, namely, main, spatial attention, and channel-wise attention branches. The main branch takes high-resolution feature maps as input to extract spatial details around boundaries. The other two attention branches boost the discriminative features in the main branch regarding the spatial and channel-wise dimensions. By fusing the branch outputs, robust features with detailed spatial information can be extracted, and accurate skin lesion boundaries can be obtained.

RESULTS

Experiments on datasets from the International Symposium on Biomedical Imaging in 2016 and 2017 and the PH2 dataset retrieved Jaccard indices of 0.783, 0.858, and 0.857, respectively, for the proposed method. Hence, our method can accurately extract skin lesion boundaries and is robust to hair fibers and artifacts in the images. Overall, our method outperforms two typical segmentation networks (FCN-8 s and U-Net) and other state-of-the-art skin lesion segmentation methods.

CONCLUSIONS

The proposed network endowed with high-resolution feature blocks preserves spatial details during feature extraction, and its attention mechanism enhances representative features while suppressing noise. Hence, the proposed approach provides high-performance skin lesion segmentation.

Bi-directional Dermoscopic Feature Learning and Multi-scale Consistent Decision Fusion for Skin Lesion Segmentation

Accurate segmentation of skin lesion from dermoscopic images is a crucial part of computer-aided diagnosis of melanoma. It is challenging due to the fact that dermoscopic images from different patients have non-negligible lesion variation, which causes difficulties in anatomical structure learning and consistent skin lesion delineation. In this paper, we propose a novel bi-directional dermoscopic feature learning (biDFL) framework to model the complex correlation between skin lesions and their informative context. By controlling feature information passing through two complementary directions, a substantially rich and discriminative feature representation is achieved. Specifically, we place biDFL module on the top of a CNN network to enhance high-level parsing performance. Furthermore, we propose a multi-scale consistent decision fusion (mCDF) that is capable of selectively focusing on the informative decisions generated from multiple classification layers. By analysis of the consistency of the decision at each position, mCDF automatically adjusts the reliability of decisions and thus allows a more insightful skin lesion delineation. The comprehensive experimental results show the effectiveness of the proposed method on skin lesion segmentation, achieving state-of-the-art performance consistently on two publicly available dermoscopic image databases.