Page 73 - 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

Although higher false-positive rates may increase clinician
workload, in cancer diagnostics, prioritising recall to minimise
missed cases is generally acceptable and clinically preferable.
Table 10 elaborates on pixel-wise classification accuracy per
class, further reinforcing the model’s robust classification
performance. Superficial-Intermediate cells recorded the highest
pixel accuracy (0.9607), closely followed by Parabasal cells
(0.8823). Such high accuracies reflect the model’s strong capacity
to correctly classify individual pixels in these clinically significant
categories. The accuracy for Dyskeratotic, Koilocytotic, and
Metaplastic classes (0.7785, 0.6685, and 0.8291, respectively),
though comparatively lower, still indicates good overall predictive
performance and aligns consistently with recall metrics.

Table 11: Pixel-level classification accuracy by cervical cell type.

S.no Cervical Cancer types Accuracy
1. Dyskeratotic 0.7785
2. Koilocytotic 0.6685
3. Metaplastic 0.8291
4. Parabasal 0.8823
5. Superficial-Intermediate 0.9607
6. Mean Pixel Accuracy 0.8238

Figure 26: Pixel accuracy for each foreground cervical cell classes

57