Less Is More: A Leakage-Controlled Study of Dermoscopic Preprocessing for Joint Skin Lesion Classification and Segmentation with YOLO26

Truong Viet Vu; Nguyen Chi Hai; Nguyen Phuc Nguyen; Ngo Hoang Tu; Vo Nguyen Quoc Bao; Nguyen Thai Anh

cs.LGOct 6, 2026

Less Is More: A Leakage-Controlled Study of Dermoscopic Preprocessing for Joint Skin Lesion Classification and Segmentation with YOLO26

Authors: Truong Viet Vu, Nguyen Chi Hai, Nguyen Phuc Nguyen, Ngo Hoang Tu, Vo Nguyen Quoc Bao, Nguyen Thai Anh

Organizations: Faculty of Information Technology, Van Lang School of Technology, Van Lang University, Ho Chi Minh City, Vietnam

Abstract

Handcrafted preprocessing is widely employed in automated dermoscopic analysis to suppress imaging artifacts and enhance lesion visibility. Nevertheless, its actual contribution to modern real-time models remains unclear, particularly when evaluation protocols do not adequately control correlations among images of the same lesion. This study presents a leakage-controlled, lesion-disjoint evaluation of dermoscopic preprocessing and augmentation for joint multi-class lesion classification and instance segmentation using a fixed nano-scale YOLO26 segmentation model (YOLO26n-seg). From HAM10000 (10,015 images), quality control yields 10,013 valid image-mask pairs from 7,468 unique lesions, partitioned into mutually exclusive sets by lesion identity. With the architecture, resolution, training budget, and evaluation protocol held fixed, we compare minimally processed images plus online augmentation against offline class balancing, DullRazor-CLAHE preprocessing, and raw-processed hybrid views, over three random seeds. On the lesion-disjoint test set, the raw baseline achieves a mask mAP $_{50:95}$ of $0.5636 \pm 0.0234$ , a Dice score of $0.9356 \pm 0.0024$ , and a macro-F1 score of $0.6917 \pm 0.0202$ . Offline augmentation does not improve the mean performance, while the combined and hybrid strategies reduce both class-aware segmentation and classification accuracy. At only 2.69 million parameters, the model runs at approximately 50 frames per second. Under a leakage-controlled, lesion-disjoint protocol with all non-input factors held fixed, minimally processed dermoscopic images combined with standard online augmentation deliver a better accuracy-efficiency trade-off than increasingly complex deterministic preprocessing, which yields no consistent joint benefit across three seeds on HAM10000.

Figures & tables

Class	Train	Validation	Test	Total
akiec	229	49	49	327
bcc	360	77	77	514
bkl	768	165	165	1,098
df	80	18	17	115
mel	778	167	167	1,112
nv	4,693	1,006	1,006	6,705

Table I : Class distribution of the lesion-disjoint dataset.

Figure 1 : Overview of the experimental pipeline.

Configuration	Training views	Validation and test views	Offline augmentation
Raw + online augmentation	Raw image followed by label-safe letterboxing	Raw image followed by the same letterboxing	None
Offline class balancing	Baseline view plus three generated variants for every non- nv training image	Identical to the raw baseline	Train only
DullRazor–CLAHE	DullRazor hair removal followed by CIELAB CLAHE and letterboxing	The same deterministic preprocessing pipeline	None
Raw–processed hybrid	One raw view and one ROI–DullRazor–Gray World view for each training image	The minimally processed raw view	None

Table II : Principal preprocessing and augmentation configurations.

Input strategy	Acc.	BAcc.	Macro-F1	Dice	IoU	$\mathrm{BF}_{1}$	Mask $\mathrm{mAP}_{50}$	Mask $\mathrm{mAP}_{50:95}$
Raw + online augmentation	$\mathbf{0.8210{\pm}0.0111}$	$\mathbf{0.6672{\pm}0.0203}$	$\mathbf{0.6917{\pm}0.0202}$	$0.9356{\pm}0.0024$	$0.8971{\pm}0.0027$	$\mathbf{0.4922{\pm}0.0048}$	$\mathbf{0.7364{\pm}0.0232}$	$\mathbf{0.5636{\pm}0.0234}$
Raw without online augmentation	$0.7753{\pm}0.0070$	$0.5707{\pm}0.0373$	$0.5893{\pm}0.0279$	$0.9228{\pm}0.0006$	$0.8815{\pm}0.0008$	$0.4582{\pm}0.0008$	$0.5654{\pm}0.0366$	$0.4271{\pm}0.0252$
Offline class balancing	$0.8150{\pm}0.0047$	$0.6654{\pm}0.0154$	$0.6870{\pm}0.0251$	$0.9332{\pm}0.0021$	$0.8943{\pm}0.0020$	$0.4841{\pm}0.0051$	$0.7305{\pm}0.0168$	$0.5614{\pm}0.0131$
DullRazor–CLAHE	$0.8026{\pm}0.0010$	$0.6090{\pm}0.0147$	$0.6356{\pm}0.0129$	$0.9304{\pm}0.0036$	$0.8900{\pm}0.0038$	$0.4713{\pm}0.0021$	$0.6940{\pm}0.0133$	$0.5192{\pm}0.0122$
Raw–processed hybrid	$0.8186{\pm}0.0058$	$0.6252{\pm}0.0161$	$0.6537{\pm}0.0257$	$\mathbf{0.9366{\pm}0.0006}$	$\mathbf{0.8978{\pm}0.0006}$	$0.4846{\pm}0.0050$	$0.7119{\pm}0.0089$	$0.5466{\pm}0.0074$

Table III : Three-seed test performance of the main YOLO26n-seg configurations. Acc., BAcc., and BF 1 denote accuracy, balanced accuracy, and boundary F 1 score, respectively; bold values indicate the best mean.

Figure 2 : Main test-set comparison of the four principal input strategies and the deterministic no-augmentation control. Points denote three-seed means, horizontal error bars denote standard deviations, and dashed lines mark the raw baseline (raw + online augmentation).

Configuration	Acc.	BAcc.	Macro-F1	Dice	IoU	$\mathrm{BF}_{1}$	Mask $\mathrm{mAP}_{50}$	Mask $\mathrm{mAP}_{50:95}$
Raw (seed 42)	$\mathbf{0.8310}$	$0.6489$	$0.6796$	$0.9331$	$0.8943$	$0.4869$	$0.7340$	$0.5520$
DullRazor only	$0.8097$	$0.6020$	$0.6440$	$0.9329$	$0.8927$	$0.4821$	$0.6967$	$0.5224$
CLAHE only	$0.8150$	$0.6505$	$0.6623$	$0.9315$	$0.8927$	$0.4846$	$0.7366$	$0.5664$
Gray World only	$0.8064$	$0.6499$	$0.6730$	$\mathbf{0.9371}$	$\mathbf{0.8987}$	$\mathbf{0.4940}$	$0.7261$	$0.5605$
ROI only	$0.8263$	$\mathbf{0.6524}$	$\mathbf{0.6945}$	$0.9341$	$0.8962$	$0.4934$	$\mathbf{0.7405}$	$\mathbf{0.5729}$
ROI + DullRazor + CLAHE	$0.8097$	$0.6310$	$0.6580$	$0.9303$	$0.8902$	$0.4736$	$0.6942$	$0.5123$

Table IV : Component ablation on the lesion-disjoint test set using seed 42. The raw reference is the single-seed counterpart of the three-seed raw baseline in Table III . Bold values indicate the best result in each column.

Architecture	Params (M)	Size (MiB)	Macro-F1	Mask $\mathrm{mAP}_{50:95}$	Latency (ms)	FPS
YOLO11n-seg	2.84	5.76	$0.6734{\pm}0.0068$	$0.5501{\pm}0.0150$	$25.51{\pm}4.39$	$39.92{\pm}6.28$
YOLO12n-seg	2.78	5.74	$0.6962{\pm}0.0178$	$0.5758{\pm}0.0067$	$20.17{\pm}6.65$	$54.33{\pm}21.76$
YOLO26n-seg	2.69	6.28	$0.6917{\pm}0.0202$	$0.5636{\pm}0.0234$	$20.48{\pm}4.24$	$50.14{\pm}9.42$
YOLO26s-seg	10.37	22.29	$0.7239$	$0.5945$	$14.26$	$70.13$

Table V : Raw-input architecture and efficiency comparison. M and MiB denote millions of parameters and mebibytes; results are mean

\pm

standard deviation over three seeds, and YOLO26s-seg is reported for seed 42. FPS is the per-seed mean of ( 5 ), which differs slightly from

1000/\bar{\ell}_{\mathrm{ms}}

as the reciprocal and mean do not commute.

Figure 3 : Qualitative test predictions of the raw YOLO26n-seg baseline using seed 52. The upper examples are correctly classified, whereas the lower examples retain high mask overlap but receive incorrect diagnostic labels.

Explore similar work

Jun 17, 2026cs.CV

PEFT-MedSAM: Efficient Fine-Tuning of Medical Foundation Models for Explainable Skin Lesion Segmentation

Automated segmentation of skin lesions using deep learning models for dermoscopic images can be very helpful in finding melanomas earlier than they would normally be detected. However, most deep learning methods available do not perform well. The aim of this paper is to present a parameter-efficient fine-tuning method called PEFT-MedSAM for adapting the Medical Segment Anything Model (MedSAM) to automatically segment dermoscopic skin lesions. The PEFT-MedSAM method uses only the lightweight mask decoder for training the model while keeping the pre-trained image encoder and prompt encoder frozen. The experiments performed on the ISIC 2018 benchmark dataset shows that PEFT-MedSAM obtains a dice coefficient of .9411 and an intersection over union value of .8918 when compared to both a fully trained U-Net baseline (.8715 dice coefficient) and zero-shot MedSAM inference (.8997 dice coefficient). The external validation of the model using PH2 dataset shows .9467 dice coefficient with +/- .0310 standard deviation. Supportive evidence for these claims include a p-value less than .0001 for Wilcoxon signed rank tests comparing the two datasets and bootstrap-estimated 95% confidence intervals of [.9364,.9447] that represent the estimated range of possible values for the average dice coefficient obtained by repeating the test. To increase clinical trustworthiness, we used Grad-CAM explainability along with a pointing game based evaluation methodology to evaluate the CNN baseline model on the validation set. The results showed that we had an accuracy rate of 98.27% on the validation set of 519 images and confirmed that the model classified regions containing skin lesions.

Asad Channa, Abdullah Khan, Asghar Ali Chandio +4

Department of Computer Science, Quaid-e-Awam University of Engineering, Sciences & Technology · Department of Artificial Intelligence, Quaid-e-Awam University of Engineering, Sciences & Technology, Pakistan · Department of Computer Science, Sindh Madressatul Islam University, City Campus, Karachi +1

Skin Lesion Classification Lesion Segmentation Melanoma Deep Learning Model Fine-Tuning

Ask PDF

Jul 21, 2026cs.CV

Privileged Lesion-Context Relational Distillation for Mask-Free Skin Lesion Classification

Accurate skin lesion classification can benefit from lesion segmentation masks, but requiring masks or an auxiliary segmentation model during inference reduces clinical practicality and increases computational complexity. This work introduces Privileged Lesion-Context Relational Distillation (PLCRD), a teacher-student framework that exploits lesion masks exclusively during training while preserving image-only inference. The privileged teacher jointly analyzes the original dermoscopic image and its mask-guided lesion region to learn lesion-specific and contextual diagnostic representations. An image-only student is then trained through complementary knowledge-transfer mechanisms that convey the teacher's diagnostic distribution, lesion-focused attention, inter-lesion relational geometry, and lesion-context structure. PLCRD decomposes deep representations into lesion and contextual embeddings and transfers their relational organization through inter-lesion similarity alignment, lesion-context affinity matching, separation regularization, and class-aware relational learning. This formulation avoids direct feature matching between heterogeneous teacher and student architectures and enables the student to internalize mask-informed diagnostic structure without accessing masks at deployment. The framework was evaluated on HAM10000 using lesion-disjoint data partitioning and externally validated on ISIC 2018 without retraining. PLCRD achieved a lesion-level macro-F1 of 0.773 +/- 0.018, balanced accuracy of 0.764 +/- 0.023, and macro-AUROC of 0.976 +/- 0.002 on HAM10000, together with a macro-F1 of 0.732 +/- 0.008 on ISIC 2018. The results indicate that privileged lesion annotations can be transformed into transferable relational knowledge, yielding a practical and interpretable approach to mask-free skin lesion classification.

Abu Mukaddim Rahi, Md Mithun Hossain, Md Zulficar Hasan Joy +2

Department of IT in Artificial Intelligence, Victoria University, Footscray, VIC, Australia · BUBT Resarch Graduate School, Bangladesh University of Business and Technology, Dhaka, Bangladesh · Department of Artificial Intelligence, Daffodil Institute of Information Technology, Dhaka, Bangladesh +2

Skin Lesion Classification Lesion Segmentation Lesion Dataset Distillation

Ask PDF

Jul 29, 2026cs.CV

Searching for Robust Augmentations to Improve Out-of-Domain Generalization in Dermoscopic Skin Cancer Classification

Background/Objectives: Dermoscopic skin-lesion classifiers lose accuracy when images arrive from a new clinic or a new device. We asked which data augmentations reduce that loss, and measured the effect under a protocol that keeps policy selection separate from policy evaluation. Methods: A ConvNeXt-Large binary malignant-versus-non-malignant classifier was trained on six dermoscopic sources (25,903 images); HAM10000 and ISIC 2016-2020 were held out of training entirely. Single augmentations, photometric combinations and eleven composite policies were ranked on a development split of 1511 held-out images. The winning policy was then evaluated on a confirmation set of 8073 held-out images that took no part in that ranking and from which we removed every image sharing a lesion identifier with the training data and every image contributed by an institution represented in training. Both policies were retrained with four random seeds each and compared with an exact permutation test. Results: The mix policy raised confirmation-set ROC-AUC from 0.787 to 0.826 (+0.039; per-seed ranges 0.772-0.797 and 0.815-0.840, non-overlapping; exact permutation p=0.029), with the same direction on each contributing source. At matched sensitivity the gain is larger in clinical terms: specificity rose from 0.612 to 0.713 at a sensitivity of 0.80, and from 0.284 to 0.397 at a sensitivity of 0.95. In-domain ROC-AUC was preserved (0.938 to 0.941). On an independent clinical cohort acquired with a different device at a different institution (472 images, 22 malignant), performance was maintained (0.934 versus 0.930). Conclusions: Augmentations that model the physical causes of domain shift improve cross-source transfer at no cost to in-domain accuracy, and the improvement survives a selection-disjoint, contamination-free evaluation.

Alexander Kozachok, Ilya Latyshev, Evgeny Karpulevich +3

Trusted AI Research Center, Russian Academy of Sciences, 109004 Moscow, Russia

Skin Lesion Classification Data Augmentation Medical Image Classification Lesion Roc-Auc

Ask PDF