Publications

Manuscripts, preprints, and research contributions.

Manuscripts Under Review

Generalizable Ensemble Deep Learning for Skin Lesion Classification:
Internal and External Validation on HAM10000 and ISIC 2019

Status: Under review (2026)
Author: Md Naim Hassan Saykat

This manuscript proposes a robust multi-model ensemble framework for automated skin lesion classification. The study integrates CNNs, ResNet, DenseNet, EfficientNet, ConvNeXt, MobileNet, and Vision Transformers, and evaluates generalization through internal validation on HAM10000 and external validation on ISIC 2019. Explainability is addressed using Grad-CAM and uncertainty-aware analysis.

  • Multi-model ensemble learning
  • Internal & external validation protocol
  • Explainable AI (Grad-CAM)
  • Medical image robustness and generalization

Preprints

A preprint of this work is publicly available on Zenodo. Additional manuscripts are currently under review and in preparation.


Open-Source Research Artifacts

All experiments, implementations, and reproducibility materials are publicly available. See the Repositories page for curated research code and datasets.


In Preparation

Edge-Ready Skin Lesion Classification with Efficient Deep Ensembles and Latency-Aware Design: A Web-Based Diagnostic Tool

Manuscript in preparation. Target submission: 2026–27 academic year

This work investigates edge-efficient and latency-aware deep learning for medical image classification, with a focus on deployable skin lesion diagnosis. Building on prior ensemble learning results on HAM10000 and ISIC 2019, the study integrates model compression, quantization, and inference benchmarking to achieve clinically meaningful performance under real-world computational constraints.

The proposed system combines efficient deep ensembles (CNN, ResNet, Vision Transformer) with Grad-CAM-based explainability and evaluates accuracy–latency trade-offs on edge-oriented platforms. A lightweight web-based diagnostic interface is developed to demonstrate practical deployment feasibility.


Note: This publication list will be updated as manuscripts progress through review and acceptance.