arXiv subjects

Jason Hagerty

Publications and source records attributed to Jason Hagerty.

3 recordsLinked to original sources

arXiv2026

Toward Artifact-Free Dermoscopy: Characterization of ISIC 2019 Image Defects for Deep Learning Applications

The International Skin Imaging Collaboration (ISIC) archive offers the largest publicly available collection of dermoscopic images for skin lesion analysis. However, the ISIC 2019 dataset suffers from several acquisition artifacts such as hair occlusions, vignetting, surgical ink markings, and ruler scales, all of which compromise its utility in deep learning workflows by introducing non-biological visual structures. To address these limitations, we present a rigorously annotated version of ISIC 2019 that includes binary masks for key artifact types. Ink marks and vignetting were annotated through systematic manual review and iterative refinement using professional image editing software. For hair and ruler artifacts, we implemented a semi-automatic pipeline based on our previously developed deep learning architecture, ChimeraNet. Automated predictions were then manually corrected as needed for precision. The resulting dataset supports both the development of our own melanoma detection model and broader community research requiring artifact-mitigated dermoscopic images.

eess.IV↗

arXiv2025

Cepstrum-Based Texture Features for Melanoma Detection

This paper introduces a set of cepstrum-based texture features for melanoma classification and validates their performance on dermoscopic images from the ISIC 2019 dataset. We propose applying gray-level co-occurrence matrix (GLCM) statistics to 2D cepstral representations, a novel approach in image analysis. Combined with established handcrafted lesion descriptors, these features were evaluated using XGBoost models. Incorporating select cepstral features improved the area under the receiver operating characteristic curve, accuracy, and F1 score for binary melanoma vs. nevus classification. Results suggest that cepstral GLCM features offer complementary discriminatory information for melanoma detection.

eess.IV↗

arXiv2023

Increasing Melanoma Diagnostic Confidence: Forcing the Convolutional Network to Learn from the Lesion

Deep learning implemented with convolutional network architectures can exceed specialists' diagnostic accuracy. However, whole-image deep learning trained on a given dataset may not generalize to other datasets. The problem arises because extra-lesional features - ruler marks, ink marks, and other melanoma correlates - may serve as information leaks. These extra-lesional features, discoverable by heat maps, degrade melanoma diagnostic performance and cause techniques learned on one data set to fail to generalize. We propose a novel technique to improve melanoma recognition by an EfficientNet model. The model trains the network to detect the lesion and learn features from the detected lesion. A generalizable elliptical segmentation model for lesions was developed, with an ellipse enclosing a lesion and the ellipse enclosed by an extended rectangle (bounding box). The minimal bounding box was extended by 20% to allow some background around the lesion. The publicly available International Skin Imaging Collaboration (ISIC) 2020 skin lesion image dataset was used to evaluate the effectiveness of the proposed method. Our test results show that the proposed method improved diagnostic accuracy by increasing the mean area under receiver operating characteristic curve (mean AUC) score from 0.9 to 0.922. Additionally, correctly diagnosed scores are also improved, providing better separation of scores, thereby increasing melanoma diagnostic confidence. The proposed lesion-focused convolutional technique warrants further study.

cs.CV↗