---
title: 'SipakMed: Cervical Cytology Dataset'
url: https://www.emergentmind.com/topics/sipakmed
type: topic
---

# SipakMed: Cervical Cytology Dataset

SipakMed—also written SIPaKMeD or SiPaKMeD in the literature—is a public Pap-smear cervical cytology dataset used for automated analysis of cervical cells. Across the cited studies, it is most commonly described as containing 4,049 manually cropped single-cell images derived from a larger set of source images and annotated into five cytomorphological categories by expert cytopathologists. The dataset has been used for five-class cell classification, grouped abnormality classification, segmentation, nuclear-center localization, and risk-scoring experiments, and it appears in work spanning CNNs, hybrid deep-feature pipelines, ensembles, vision transformers, and multi-task architectures [2108.09460][2105.07402].

## 1. Corpus, provenance, and naming conventions

Most studies describe SipakMed as a corpus of 4,049 isolated single-cell Pap-smear images together with 966 source images from which the cell crops were obtained. In the single-cell setting, the crops are already centered on individual cells, so many classification pipelines omit any explicit segmentation stage [2108.09460][2410.17735]. The dataset is also described as publicly available on Kaggle in one optimizer-comparison study [2410.17735].

The literature is not fully uniform in how it characterizes the source imagery. One study refers to 966 cluster-cell images and 4,049 manually cropped isolated-cell images [2108.09460]. A later segmentation-and-risk paper describes 966 high-resolution whole-slide images of size \(2048\times1536\), from which 4,049 single-cell crops were manually extracted [2505.15505]. Another slide-analysis paper instead reports 204 non-overlapping fields of view of size \(1280\times960\) acquired from Papanicolaou-stained slides, with the isolated-cell PNG crops exported from manually delineated ROIs [2504.20435]. This suggests that downstream papers do not always preserve a single, fully consistent description of the acquisition hierarchy.

Original crop sizes are variable rather than standardized. One comparative deep-learning study reports a range from \(62\times48\) to \(531\times553\) pixels [2402.15905], while CerviFormer reports a range from roughly \(71\times59\) to \(490\times474\) pixels [2303.10222]. A robust-comparison paper emphasizes that this variability carries diagnostically relevant shape information, especially for cytomorphological cues such as the nucleus-to-cytoplasm ratio [2105.07402].

## 2. Cytomorphological taxonomy and annotation structure

The most frequently repeated five-class taxonomy in the cited literature is shown below.

| Class | Count | Interpretation reported in the literature |
|---|---:|---|
| Superficial-Intermediate | 831 | Mature squamous cells; normally predominant |
| Parabasal | 787 | Small immature basal-like cells |
| Koilocytotic | 825 | HPV-infected cells with perinuclear clearing |
| Dyskeratotic | 813 | Premature keratinization; associated with atypia |
| Metaplastic | 793 | Cells undergoing squamous metaplasia |

This distribution is reported consistently in multiple studies using the single-cell subset [2105.07402][2106.04919][2108.09460][2308.02781][2102.12191]. The labels are morphological rather than purely diagnostic: two classes are commonly treated as normal, two as abnormal, and metaplastic as benign in grouped-class experiments. CerviFormer, for example, pools parabasal and superficial-intermediate into “Normal” (1,618 images), dyskeratotic and koilocytotic into “Abnormal” (1,638 images), and keeps metaplastic as “Benign” (793 images) for three-state classification [2303.10222].

Ground-truth labels are generally inherited from the released dataset rather than re-annotated by later authors. One transfer-learning study explicitly states that label assignment was inherited from the Kaggle release and that the authors did not perform additional relabeling, re-annotation, or expert-consensus procedures [2410.17735]. For segmentation-oriented work, additional annotation layers are reported. A Dense-UNet201 study states that ground-truth masks delineating nucleus and cytoplasm were available as \(*.dat\) annotation files and were decoded into binary mask images [2504.12807]. A different multi-task segmentation study reports patch-level binary masks for source images, with white denoting cell and black denoting background [2505.15505].

Not all later papers preserve the same class ontology. Two 2025 studies relabel the five categories as superficial squamous epithelial cells, intermediate squamous epithelial cells, columnar epithelial cells, low-grade squamous intraepithelial lesion cells, and high-grade squamous intraepithelial lesion cells, or very close variants of that scheme [2504.12807][2506.15489]. Another 2024 study reports a different class breakdown, including Parabasal = 727 and Superficial-Intermediate = 813 [2402.15905]. This suggests that any cross-paper comparison should verify the exact label mapping used in the reported experiment.

## 3. Image geometry, preprocessing, and aspect-ratio handling

A central practical issue in SipakMed usage is normalization of highly variable cell-crop sizes to fixed network inputs. The most common target resolution is \(224\times224\) pixels, used in transfer-learning and large-model comparison studies [2410.17735][2105.07402]. Other reported input sizes include \(110\times110\) with aspect-ratio preservation [2402.15905], \(192\times192\) [2408.10498], \(256\times256\) [2308.02781], \(100\times100\) [2409.06220], and \(75\times75\) [2504.20435].

Two broad preprocessing strategies recur. The first is direct resizing, which forces every crop to a fixed grid and distorts the original cell aspect ratio. The second is aspect-ratio-preserving square embedding, in which the shorter side is zero-padded before downsampling to the network input size [2105.07402]. The systematic comparison in that study found that 12 of 22 models improved under direct resize and 10 decreased, with an average Scaled-minus-Standard accuracy change of \(-0.31\%\pm1.41\%\), which the authors interpret as negligible [2105.07402]. Their account attributes this robustness to the persistence of nuclear size, chromatin texture, and boundary detail even when whole-cell shape cues are distorted.

Normalization and augmentation practices vary substantially. Some studies report only resizing and generic “image augmentation,” without separate intensity normalization, color standardization, denoising, or filtering [2410.17735]. Others use ImageNet mean-standard-deviation normalization [2108.09460], zero-mean unit-variance normalization [2511.03645], or domain-specific enhancement such as Perona-Malik diffusion and CLAHE [2506.15489]. Augmentation ranges from none explicitly reported [2108.09460] to extensive affine, photometric, noise-injection, zoom, padding, and contrast perturbations [2402.15905][2102.12191].

Because the single-cell subset is already pre-cropped, many classification studies treat SipakMed as a direct recognition problem rather than a detect-then-classify problem [2410.17735]. By contrast, localization work derives a nuclear centroid from contour annotations and then pads each image to a \(256\times256\) frame for coordinate regression [2511.03645].

## 4. Experimental protocols and task formulations

Evaluation on SipakMed is protocol-dependent. Reported splits include \(60\%/20\%/20\%\) train/validation/test [2105.07402][2108.09460], \(80\%/10\%/10\%\) [2410.17735][2506.15489], \(80\%/20\%\) train/test without a distinct validation set [2408.10498], \(70\%/20\%/10\%\) on an augmented dataset [2409.06220], and five-fold cross-validation [2508.11181][2504.20435][2511.03645]. Reported metrics therefore are not directly commensurate across studies.

The dataset supports several task formulations.

| Task | Representative formulation | Representative metrics |
|---|---|---|
| Single-cell classification | 5-class, 3-class, or 2-class | Accuracy, precision, recall, \(F_1\), loss, AUC |
| Segmentation | Pixel-wise nucleus/cytoplasm or cell/background masks | IoU, Dice, accuracy, loss |
| Localization | Nuclear-center regression from contour centroid | \(R^2\), MSE |
| Risk scoring | Class-conditional likelihoods from learned features | Posterior probabilities, cosine similarity |

Five-class classification is the dominant use case, but grouped-label settings are common. DeepCervix reports 2-class, 3-class, and 5-class experiments on the same base dataset [2102.12191]. CerviFormer uses the three-state Normal/Abnormal/Benign grouping [2303.10222]. Whole-slide or cluster-image classification also appears in the Sugeno-fusion study, which treats both the 966-image WSI subset and the 4,049-image SCI subset as separate benchmarks [2108.09460].

Segmentation formulations are less standardized. One paper uses resized \(256\times256\) RGB images and corresponding binary JPG masks decoded from \(*_cyt*.dat\) files [2504.12807]. Another uses \(512\times512\) overlapping patches extracted from source images, discarding all-black patches and learning a multi-task U-Net for simultaneous segmentation and classification [2505.15505]. A localization study instead treats each sample as an image-annotation pair and regresses the centroid \((h_{\rm gt},k_{\rm gt})\) of the annotated nuclear contour [2511.03645].

## 5. Benchmark results and methodological findings

Reported performance on SipakMed spans a wide range, and the range itself is informative because it reflects changes in split protocol, label grouping, preprocessing, augmentation, and model family. Under an \(80/10/10\) protocol with ResNet-18, ResNet-34, and VGG-16, one transfer-learning study reports best accuracies of \(72.8\%\) for VGG-16 with Adamax, \(66.8\%\) for ResNet-18 with Adamax, and \(54.0\%\) for ResNet-34, with Nadam exceeding Adamax on ResNet-34 by \(0.034\%\) [2410.17735]. In the same study, the central conclusion is that transfer learning improves all CNNs and optimizers and that optimizer choice has little influence once transfer learning is used [2410.17735].

On a \(60/20/20\) split across 22 deep models, the aspect-ratio comparison study reports that GoogLeNet achieved \(96.03\%\) accuracy on directly resized data, DeiT \(95.42\%\), DenseNet169 \(95.17\%\) on aspect-ratio-preserving standard data, and DeiT \(94.18\%\) on standard data [2105.07402]. Cost-sensitive and supervised-contrastive training on the single-cell subset yields \(97.29\%\) accuracy, with the paper reporting that class weighting modestly improved minority-class recall and that contrastive pre-training improved inter-class separability, especially between dyskeratotic and koilocytotic cells [2402.15905]. A hybrid pipeline based on deep feature extraction, PCA preserving \(99\%\) variance, Grey Wolf Optimizer feature selection, and RBF-SVM reports \(97.87\%\) test accuracy [2106.04919].

Ensemble methods push reported accuracy higher under their own protocols. A Sugeno fuzzy-integral fusion of Inception v3, DenseNet-161, and ResNet-34 reaches \(98.54\%\) accuracy on the SCI subset and \(96.33\%\) on the WSI subset [2108.09460]. A Voting-Stacking ensemble of three Inception-family models reports \(99.75\%\), \(99.51\%\), and \(99.34\%\) accuracy under \(10\%\), \(20\%\), and \(30\%\) test hold-outs, with a three-level super-ensemble reaching \(100\%\) in the \(10\%\)-hold-out setting [2308.02781]. DeepCervix, which fuses \(1024\)-dimensional pre-softmax features from VGG16, VGG19, ResNet50, and Xception into a \(4096\)-dimensional representation, reports \(99.85\%\) for 2-class, \(99.38\%\) for 3-class, and \(99.14\%\) for 5-class classification [2102.12191].

Transformer-based approaches show similarly heterogeneous but competitive behavior. CerviFormer applies cross-attention and a latent-transformer bottleneck to \(224\times224\) images partitioned into \(14\times14\) patches and reports \(93.70\%\) accuracy for 3-state classification, along with \(92.99\%\) sensitivity, \(96.81\%\) specificity, and Cohen’s \(\kappa=0.90\) [2303.10222]. HistoViT fine-tunes a ViT-Base patch16 model and reports \(96.94\%\) accuracy, \(96.96\%\) mean precision, \(96.94\%\) mean recall, and \(99.83\%\) AUC under five-fold cross-validation [2508.11181]. A CvT-13 classifier trained for low-resource slide analysis reports \(99.68\%\) accuracy and average AUC \(0.9997\) under five-fold cross-validation when fine-tuned from ImageNet-22k [2504.20435].

SipakMed has also supported non-classification tasks. A multi-task U-Net study reports IoU \(=0.83\) for segmentation and \(90\%\) accuracy for simultaneous classification, while its standalone Multi-Resolution Fusion Deep Convolutional Network reaches \(91.28\%\) classification accuracy using about \(1.7\) million parameters [2505.15505]. A Dense-UNet201 plus Spider Monkey Optimization study reports segmentation accuracy \(96.16\%\), IoU \(91.63\%\), and Dice \(95.63\%\) [2504.12807]. For nuclear-center localization, replacing pure CoordConv channels with intensity-weighted coordinate channels improves mean test-set \(R^2\) by \(0.02898\) with a \(95\%\) bootstrap CI of \([0.02362,\,0.03349]\) and one-sided \(p<5\times10^{-5}\) [2511.03645].

## 6. Limitations, inconsistencies, and research significance

Several constraints recur across SipakMed-based studies. The total single-cell sample count of 4,049 is small for end-to-end deep learning, which is why transfer learning, augmentation, and feature reuse appear so frequently [2410.17735]. The dataset is fundamentally a single-cell crop collection; one study explicitly notes that no clinical metadata or multi-cell fields of view are integrated and that the dataset is “strictly single-cell crops” [2410.17735]. A plausible implication is that SipakMed is strongest as a controlled cytomorphology benchmark rather than a complete proxy for slide-level screening workflows.

The literature also exposes several interpretive pitfalls. Source-image descriptions differ, class ontologies are not always preserved, and train/validation/test protocols vary substantially [2505.15505][2506.15489]. Some papers omit details on microscope magnification, staining variants, stratification, validation protocol, or class-imbalance handling [2408.10498][2410.17735]. This makes raw leaderboard-style comparison unreliable unless preprocessing, grouping, augmentation, and evaluation schemes are aligned.

At the same time, the dataset has been unusually productive methodologically. It has been used to test whether aspect-ratio preservation matters [2105.07402], whether cost-sensitive and contrastive learning improve minority recall and representation geometry [2402.15905], whether hybrid deep features and swarm-based optimization outperform direct classifiers [2106.04919][2504.12807], and whether lightweight models can approximate larger backbones at lower computational cost [2505.15505][2409.06220]. The breadth of these uses indicates that SipakMed functions not only as a classification corpus but also as a compact experimental platform for studying preprocessing, optimization, explainability, segmentation, localization, and low-resource deployment in cervical cytology.

Source: https://www.emergentmind.com/topics/sipakmed