Skin Lesion Analysis Towards Melanoma Detection Using Deep Learning Network (1703.00577v2)

Abstract: Skin lesion is a severe disease in world-wide extent. Early detection of melanoma in dermoscopy images significantly increases the survival rate. However, the accurate recognition of melanoma is extremely challenging due to the following reasons, e.g. low contrast between lesions and skin, visual similarity between melanoma and non-melanoma lesions, etc. Hence, reliable automatic detection of skin tumors is very useful to increase the accuracy and efficiency of pathologists. International Skin Imaging Collaboration (ISIC) is a challenge focusing on the automatic analysis of skin lesion. In this paper, we proposed two deep learning methods to address all the three tasks announced in ISIC 2017, i.e. lesion segmentation (task 1), lesion dermoscopic feature extraction (task 2) and lesion classification (task 3). A deep learning framework consisting of two fully-convolutional residual networks (FCRN) is proposed to simultaneously produce the segmentation result and the coarse classification result. A lesion index calculation unit (LICU) is developed to refine the coarse classification results by calculating the distance heat-map. A straight-forward CNN is proposed for the dermoscopic feature extraction task. To our best knowledges, we are not aware of any previous work proposed for this task. The proposed deep learning frameworks were evaluated on the ISIC 2017 testing set. Experimental results show the promising accuracies of our frameworks, i.e. 0.718 for task 1, 0.833 for task 2 and 0.823 for task 3 were achieved.

An Analytical Overview of Deep Learning Approaches for Melanoma Detection

The paper "Skin Lesion Analysis towards Melanoma Detection Using Deep Learning Network" by Yuexiang Li and Linlin Shen presents a systematic exploration of deep learning frameworks applied to the analysis of dermatological images for melanoma detection. The paper specifically focuses on the tasks set forth by the ISIC 2017 challenge: lesion segmentation, dermoscopic feature extraction, and lesion classification.

Methodological Approach

The authors introduce two distinct deep learning frameworks: the Lesion Indexing Network (LIN) and the Lesion Feature Network (LFN). Each is tailored to address the unique challenges presented by dermatological image analysis.

1. Lesion Indexing Network (LIN): This framework employs fully-convolutional residual networks (FCRN-88) to simultaneously execute lesion segmentation and classification tasks. The LIN utilizes multiple scales of input images to leverage varying resolutions, thereby refining the model's performance. A distinctive component, the Lesion Index Calculation Unit (LICU), is integrated to assess pixel importance for lesion classification. This component optimizes the classification probabilities derived from the FCRNs, thereby improving the accuracy of the predictions.
2. Lesion Feature Network (LFN): This CNN-based framework is designed specifically for the extraction of dermoscopic features. Training involves patches derived from superpixel masks of dermoscopic images. The LFN's architecture is tested against variations in network depth and feature representation breadth, examining the impact of techniques such as batch normalization and weighted softmax loss.

Experimental Results and Numerical Outcomes

During evaluation on the ISIC 2017 testing set, the LIN framework achieved a Jaccard Index (JA) of 0.718 for segmentation and an AUC of 0.823 for classification. Notably, the LFN yielded impressive numbers in dermoscopic feature extraction, with an area under the curve (AUC) of 0.833. Notably, the approaches used here demonstrated competitive performance compared to other state-of-the-art methods.

Implications and Future Prospects

The dual implementation of LIN and LFN offers valuable insights into the potential of deep learning frameworks in medical image processing. The experimentation with multi-scale inputs, network architectures, and augmentation strategies lays a solid groundwork for further research into automated melanoma detection. Moreover, no apparent prior work has established benchmarks for the dermoscopic feature extraction task, suggesting that the LFN's results can serve as a reference point for subsequent studies.

Conclusion

The paper enhances understanding of how deep learning methodologies can be applied to detect melanoma effectively and efficiently. The insights gained point towards the increasing applicability of AI-driven approaches in the medical field, potentially influencing future research directions in developing robust, real-time diagnostic systems. The frameworks presented by Li and Shen contribute meaningfully to ongoing efforts in improving automatic detection systems for dermatological analysis.