[Paper Review] Automated Skin Lesion Classification Using Ensemble of Deep Neural Networks in ISIC 2018: Skin Lesion Analysis Towards Melanoma Detection Challenge

Ashraful Alam Milton|arXiv (Cornell University)|Jan 30, 2019

Cutaneous Melanoma Detection and Management11 references82 citations

TL;DR

The paper evaluates multiple deep neural networks for melanoma detection on the ISIC 2018 dataset and reports the best validation score achieved by PNASNet-5-Large within an ensemble framework.

ABSTRACT

In this paper, we studied extensively on different deep learning based methods to detect melanoma and skin lesion cancers. Melanoma, a form of malignant skin cancer is very threatening to health. Proper diagnosis of melanoma at an earlier stage is crucial for the success rate of complete cure. Dermoscopic images with Benign and malignant forms of skin cancer can be analyzed by computer vision system to streamline the process of skin cancer detection. In this study, we experimented with various neural networks which employ recent deep learning based models like PNASNet-5-Large, InceptionResNetV2, SENet154, InceptionV4. Dermoscopic images are properly processed and augmented before feeding them into the network. We tested our methods on International Skin Imaging Collaboration (ISIC) 2018 challenge dataset. Our system has achieved best validation score of 0.76 for PNASNet-5-Large model. Further improvement and optimization of the proposed methods with a bigger training dataset and carefully chosen hyper-parameter could improve the performances. The code available for download at https://github.com/miltonbd/ISIC_2018_classification

Motivation & Objective

• Motivate automated melanoma detection to aid early diagnosis and clinical decision-making.
• Assess the performance of several deep learning architectures on the ISIC 2018 challenge dataset.
• Explore data processing and augmentation strategies to improve dermoscopic image classification.
• Identify the best-performing model within an ensemble approach for skin lesion classification.

Proposed method

• Experiment with deep networks such as PNASNet-5-Large, InceptionResNetV2, SENet154, and InceptionV4 on the ISIC 2018 dataset.
• Properly preprocess and augment dermoscopic images before network training.
• Evaluate models on the ISIC 2018 validation set and report performance metrics.
• Highlight that the best validation score achieved is 0.76 for PNASNet-5-Large.
• Suggest that further improvements may come from larger training data and careful hyper-parameter selection.

Experimental results

Research questions

• RQ1 Can deep neural network ensembles improve melanoma detection accuracy on the ISIC 2018 dataset?
• RQ2 Which network architectures offer the best validation performance for dermoscopic skin lesion classification?
• RQ3 How do data preprocessing and augmentation influence model performance on ISIC 2018?
• RQ4 What is the potential of larger training data and optimized hyper-parameters to further boost results?

Key findings

• PNASNet-5-Large achieved the best validation score of 0.76 among tested models.
• Various deep learning models (InceptionResNetV2, SENet154, InceptionV4) were evaluated on ISIC 2018.
• Dermoscopic images were properly processed and augmented before feeding into networks.
• The authors suggest that larger training datasets and optimized hyper-parameters could improve performance.
• Code for the approach is available at the provided URL.