Abstract
Melanoma, one of the most dangerous types of skin cancer, results in a very high mortality rate. Early detection and resection are two key points for a successful cure. Recent researches have used artificial intelligence to classify melanoma and nevus and to compare the assessment of these algorithms to that of dermatologists. However, training neural networks on an imbalanced dataset leads to imbalanced performance, the specificity is very high but the sensitivity is very low. This study proposes a method for improving melanoma prediction on an imbalanced dataset by reconstructed appropriate CNN architecture and optimized algorithms. The contributions involve three key features as custom loss function, custom mini-batch logic, and reformed fully connected layers. In the experiment, the training dataset is kept up to date including 17,302 images of melanoma and nevus which is the largest dataset by far. The model performance is compared to that of 157 dermatologists from 12 university hospitals in Germany based on the same dataset. The experimental results prove that our proposed approach outperforms all 157 dermatologists and achieves higher performance than the state-of-the-art approach with area under the curve of 94.4%, sensitivity of 85.0%, and specificity of 95.0%. Moreover, using the best threshold shows the most balanced measure compare to other researches, and is promisingly application to medical diagnosis, with sensitivity of 90.0% and specificity of 93.8%. To foster further research and allow for replicability, we made the source code and data splits of all our experiments publicly available.
Highlights
Melanoma is one of the most dangerous skin cancers, which accounts for the majority of skin cancer deaths
batch logic and loss function (BLF)’s specificity is only lower than that of ORI by 0.6 percentage points while its SEN is 4.7 percentage points higher. This demonstrates the effectiveness of BLF in maintaining high area under the curve (AUC) and SPE while balancing SEN and SPE
In terms of using the same Binary Cross-Entropy loss function, our proposed method using DenseNet[169] has an AUC 3.2% higher than InceptionV3’s method when we evaluate the models by the Test-10 dataset
Summary
Melanoma is one of the most dangerous skin cancers, which accounts for the majority of skin cancer deaths. Computer vision in the field of artificial intelligence (AI) has recently achieved very excellent results in image recognition, even exceed humans in some problems with large datasets[2] This inspires many studies about AI-based solutions for automating melanoma diagnosis using skin lesion images, especially using deep Convolution Neural Networks (CNN) for the melanoma prediction p roblem[3,4,5,6,7]. The CLF improves the learning ability of the neural networks on the minority class of melanoma It enhances the balance performance of the models which is evaluated by sensitivity (SEN) and specificity (SPE) as melanoma prediction problems. Our proposed custom mini-batch logic ensures the proportion of melanoma and nevus in every mini-batch of the training
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have