Page 61 - eBook_Proceedings of the International Conference on Digital Manufacturing V2

Fabrication and Characterization of a Low-Cost Piezoelectric using Rochelle
Salt for Energy Harvesting and Sensor Applications

Figure 20: Cervical cancer tissue cell example (Rutili de Lima,
Khan, Shah & Ferri, 2023).

In response, researchers have increasingly turned to artificial
intelligence (AI) to develop computational models capable of
analysing Pap smear images with improved accuracy and
efficiency. Early efforts in this domain were primarily driven by
traditional machine-learning techniques that relied on manual
feature extraction and handcrafted rules. Algorithms such as
support vector machines (SVMs), decision trees, and ensemble
methods were commonly used for classification. While useful as
a starting point, these models were limited in their ability to
generalise across diverse datasets due to the morphological
complexity of cervical cells (Jia, Li & Zhang, 2020). Researchers
began integrating deep learning, particularly Convolutional
Neural Networks (CNNs), to overcome these limitations in
diagnostic pipelines. Studies, such as (Pramanik et al., 2022) and
(Attallah, 2023) employed CNNs including VGG-16, CaffeNet,
and shallow custom architectures as feature extractors, feeding
their outputs into Extreme Learning Machines (ELMs) for
classification. One study implemented a two-stage ELM pipeline:
the first stage determined whether a cell was normal or abnormal,
and the second provided finer subclassification. Similarly, the
IDCNN-CDC model (Fan et al., 2023) combined Gaussian
filtering, Tsallis entropy with dragonfly optimisation (TE-DFO)

45