Evidence-Driven Differential Diagnosis of Malignant Melanoma

Naren Akash; Anirudh Kaushik; Jayanthi Sivaswamy

eess.IVAug 28, 2026

Evidence-Driven Differential Diagnosis of Malignant Melanoma

Authors: Naren Akash, Anirudh Kaushik, Jayanthi Sivaswamy

Organizations: Center for Visual Information Technology International Institute of Information Technology Hyderabad, India

Abstract

We present a modular and multi-level framework for the differential diagnosis of malignant melanoma. Our framework integrates contextual information and evidence at the lesion, patient, and population levels, enabling decision-making at each level. We introduce an anatomic-site aware masked transformer, which effectively models the patient context by considering all lesions in a patient, which can be variable in count, and their site of incidence. Additionally, we incorporate patient metadata via learnable demographics embeddings to capture population statistics. Through extensive experiments, we explore the influence of specific information on the decision-making process and examine the tradeoff in metrics when considering different types of information. Validation results using the SIIM-ISIC 2020 dataset indicate including the lesion context with location and metadata improves specificity by 17.15% and 7.14%, respectively, while enhancing balanced accuracy. The code is available at https://github.com/narenakash/meldd.

Figures & tables

Figure 1: The dermatologists’ melanoma diagnostic reasoning process (left) and the pipeline of the proposed MelDD framework (right) inspired by the clinical process. Lesion features are extracted and grouped by patients first, and lesion anatomical site information and patient metadata are incorporated later for enhanced context.

MelDD variants	PC	VC	L	M	BACC	SN	SP	AUC
V0 (Baseline)	✗	–	✗	✗	$0.7649$	0.8867	$0.6431$	$0.8371$
V1	✓	✓	✗	✗	$0.7841$	$0.8679$	$0.7003$	$0.8558$
V2	✓	✓	✓	✗	0.7904	$0.8274$	0.7534	0.8612
V3	✓	✓	✗	✓	$0.7867$	$0.8843$	$0.6890$	$0.8544$
V4	✓	✓	✓	✓	$0.7793$	$0.8761$	$0.6825$	$0.8504$
CI-Net [ 15 ]	✗	–	✗	✗	$0.6200$	$0.3220$	$0.9180$	$0.9230$

Table 1: Comparision of classification performance in melanoma recognition averaged across five-folds on SIIM-ISIC 2020 dataset: BACC: balanced accuracy, SN: sensitivity, SP: specificity at Youden’s J statistic cut-off, and ROC AUC. (PC = patient context, VC = varying lesion count, L = anatomical location, M = metadata).

MelDD variants	PC	VC	L	M	BACC	SN	SP	AUC
V0	✗	–	✗	✗	0.7649	0.8867	0.6431	0.8371
V1	✓	✓	✗	✗	$+2.51\%$	$-2.12\%$	$+8.89\%$	$+2.23\%$
V2	✓	✓	✓	✗	+3.33%	$-6.69\%$	+17.15%	+2.88%
V3	✓	✓	✗	✓	$+2.85\%$	$-0.27\%$	$+7.14\%$	$+2.07\%$
V4	✓	✓	✓	✓	$+1.88\%$	$-1.20\%$	$+6.13\%$	$+1.59\%$

Table 2: Performance improvement of the variants over the baseline (in percentage).

Figure 2: Examples of malignant melanoma prediction changes with additional context and evidence information. Green/red boxes indicate correct/incorrect predictions, respectively. In Patient A, multiple atypical lesions reduce suspicion of malignancy in an additional atypical lesion, while a morphologically typical lesion distinct in the nevus landscape is considered suspicious. Patient B demonstrates how including anatomical location accurately detects an ugly duckling suspicious lesion by comparing it to other lesions in the same location to predict malignancy effectively. The examples of Patients C and D underscore how incorporating location information prevents the misclassification of benign lesions as malignant by considering the specific anatomical characteristics that differentiate suspicious lesions in different locations. Lastly, Patients E and F emphasize the importance of patient demographics to help the model correlate lesion characteristics with susceptibility to risk factors, avoiding misdiagnosis of benign lesions as malignant based on a better understanding of patient-specific factors.

Explore similar work

DNA Methylation Profiling in Melanoma: From Lesion Classification to Therapeutic Stratification
Aug 19, 2026Jana T. Winterstein, Lukas Heinlein, Günter Raddatz +34Skin Lesion Classification Cancer Genomics
Division of Digital Prevention, Diagnostics and Therapy Guidance, German Cancer Research Center (DKFZ), INF 223, 69120 Heidelberg, Germany · Medical Faculty, University Heidelberg, Heidelberg, Germany · Department of Epigenetics, German Cancer Research Center (DKFZ), Heidelberg, Germany +24
Latent Space Analysis for Interpretable Uncertainty in Melanoma Classification
Jun 23, 2025Ciro Listone, Aniello MuranoVariational Autoencoders Skin Lesion Classification
University of Naples Federico II, Naples, Italy.
Geometry-Aware Superpixel Graph Transformer with Metadata for Skin Lesion Classification
Jun 18, 2026Muhammad Azeem, Tanveer Hussain, Amr Ahmed +1Skin Lesion Classification Graph Representation Learning
Edge Hill University, Ormskirk, Lancashire, L39 4QP, United Kingdom

Aug 19, 2026J/K move · Enter open · S save

DNA Methylation Profiling in Melanoma: From Lesion Classification to Therapeutic Stratification

Jana T. Winterstein, Lukas Heinlein, Günter Raddatz, Carina Nogueira Garcia+33

Division of Digital Prevention, Diagnostics and Therapy Guidance, German Cancer Research Center (DKFZ), INF 223, 69120 Heidelberg, Germany · Medical Faculty, University Heidelberg, Heidelberg, Germany · Department of Epigenetics, German Cancer Research Center (DKFZ), Heidelberg, Germany+24

Skin Lesion Classification Cancer Genomics

Open paper AskPDF arXiv

DNA methylation provides a stable record of cellular identity, capturing epigenetic programs that distinguish specialized cell states despite a shared genome. Because malignant transformation and tumour progression are accompanied by extensive epigenetic remodeling, we hypothesized that the methylome of melanocytic lesions contains biologically and clinically relevant information for both diagnosis and disease progression. In a cohort of 1,001 tissue samples prospectively collected across eight German university hospitals profiled using Illumina Infinium MethylationEPIC arrays, we compared machine-learning models based on selected Cytosine phosphate Guanine (CpG) methylation sites with models incorporating biology-guided features, including epigenetic age acceleration, cell type composition and copy-number variation burden. In an external test set, the best diagnostic classifier was CpG-based and distinguished melanocytic nevi, noninvasive melanoma and invasive melanoma with a macro-averaged area under the receiver operating characteristic curve of 0.919 (95% CI: 0.878 to 0.952). Notably, across CpGs most strongly hyper- and hypomethylated between NV and IM, NIM showed an intermediate methylation profile, providing a molecular correlate of its diagnostic complexity. The best model for clinically relevant treatment group prediction, with AJCC stages grouped according to guideline-based management recommendations, relied on biology-guided features and achieved a macro-averaged mean absolute error of 0.627 (95% CI: 0.477 to 0.808). Together, these findings demonstrate that methylation-based models can capture both diagnostic identity and clinically relevant disease stratification, supporting DNA methylation as a promising biomarker for further validation and potential clinical translation.

Abstract

Explore similar work

DNA Methylation Profiling in Melanoma: From Lesion Classification to Therapeutic Stratification

Latent Space Analysis for Interpretable Uncertainty in Melanoma Classification

Geometry-Aware Superpixel Graph Transformer with Metadata for Skin Lesion Classification