DeepPap: Deep Convolutional Networks for Cervical Cell Classification (1801.08616v1)

Abstract: Automation-assisted cervical screening via Pap smear or liquid-based cytology (LBC) is a highly effective cell imaging based cancer detection tool, where cells are partitioned into "abnormal" and "normal" categories. However, the success of most traditional classification methods relies on the presence of accurate cell segmentations. Despite sixty years of research in this field, accurate segmentation remains a challenge in the presence of cell clusters and pathologies. Moreover, previous classification methods are only built upon the extraction of hand-crafted features, such as morphology and texture. This paper addresses these limitations by proposing a method to directly classify cervical cells - without prior segmentation - based on deep features, using convolutional neural networks (ConvNets). First, the ConvNet is pre-trained on a natural image dataset. It is subsequently fine-tuned on a cervical cell dataset consisting of adaptively re-sampled image patches coarsely centered on the nuclei. In the testing phase, aggregation is used to average the prediction scores of a similar set of image patches. The proposed method is evaluated on both Pap smear and LBC datasets. Results show that our method outperforms previous algorithms in classification accuracy (98.3%), area under the curve (AUC) (0.99) values, and especially specificity (98.3%), when applied to the Herlev benchmark Pap smear dataset and evaluated using five-fold cross-validation. Similar superior performances are also achieved on the HEMLBC (H&E stained manual LBC) dataset. Our method is promising for the development of automation-assisted reading systems in primary cervical screening.

• The paper introduces a deep learning approach by using ConvNets pre-trained on ImageNet to eliminate the need for prior cell segmentation.
• The method achieves 98.3% accuracy and a 0.998 AUC on benchmark datasets, demonstrating superior performance in cervical cell classification.
• The study emphasizes transfer learning combined with data augmentation, paving the way for scalable and robust cervical screening in clinical settings.

Overview of DeepPap: Deep Convolutional Networks for Cervical Cell Classification

The paper "DeepPap: Deep Convolutional Networks for Cervical Cell Classification" presents a novel approach to the automation-assisted screening of cervical cytology specimens, leveraging deep learning through convolutional neural networks (ConvNets). This method directly addresses the challenges in traditional cervical cell classification methodologies, namely the dependency on accurate cell segmentation and the reliance on hand-crafted feature extraction. By executing classification without prior segmentation, DeepPap introduces a shift towards leveraging deep features, exploiting ConvNet architectures initially trained on large-scale natural image datasets.

Methodology

DeepPap employs a ConvNet model initially trained on the ImageNet dataset, subsequently fine-tuned using cervical cell datasets with image patches centered around the nuclei. Significantly, the proposed model circumvents the necessity for precise cell segmentation by relying solely on patch-based data, facilitating robust feature extraction and classification. Image augmentations such as rotations and translations are applied, ensuring variability and robustness against inaccuracies in centroid localization.

Transfer learning is a prominent aspect, where the general feature learning capabilities of models pre-trained on large, diverse datasets are harnessed to deal with data scarcity in the domain of cervical cytology. Specifically, the weights from the ConvNet layers pre-trained on ImageNet provide a foundational model, fine-tuned to optimize cervical cell classification tasks.

Results

The experimental evaluation conducted demonstrates the method’s superior performance on benchmark datasets, namely the Herlev Pap smear and HEMLBC datasets. DeepPap achieves remarkable classification accuracy of 98.3% with an AUC of 0.998 on the Herlev dataset, exemplifying the potential for high sensitivity and specificity in detecting abnormal cervical cells. Additionally, the method holds promise for practical deployment due to its resilience to moderate inaccuracies in nucleus localization, evidenced by minor performance degradation when nucleus centers are translated.

Implications and Future Directions

DeepPap provides a scalable pathway toward automation-assisted cervical screening, promising reduced labor costs and increased coverage, potentially aiding screen programs in under-resourced settings. The high specificity achieved indicates a reduced burden on cytotechnologists for follow-up examination of false positives, a common issue in wide-scale screening programs.

The continued evolution of such deep learning methodologies will likely involve integration with a broader suite of image analysis techniques to encompass overlapping cells, artifacts, and varied staining methods. Further research might focus on end-to-end pipeline optimization, from slide preparation through microscopy and image acquisition protocols to ConvNet classification, ensuring robustness across various clinical settings.

In conclusion, DeepPap showcases the effectiveness of deploying deep learning models within biomedical imaging, specifically for cytological specimen classification, indicating a transformative direction for clinical diagnostics powered by artificial intelligence. Future work should capitalize on expanding datasets and enhancing network architectures to solidify the role of AI in enhancing healthcare outcomes.