Breast cancer detection and classification: A study on the specification and implementation of multilayer perceptron analog artificial neural networks.
Journal:
Computers in biology and medicine
PMID:
40147189
Abstract
Breast cancer is a leading cause of mortality worldwide. Screening therefore remains the best defense against this disease, highlighting the need for accurate and efficient diagnostic methods. Previous authors addressed this issue by implementing digital hardware with low processing speed and high power consumption constraints. Moreover, a similar approach was developed focusing on implementing an off-chip learning on-chip inference methodology resulting in a high complexity of the proposed systems and complex communication between modules regarding the operations performed by each circuit. This research presents a fully integrated multilayer perceptron analog artificial neural network, classifying breast tumor cells as benign or malignant. A novel analog network architecture and fundamental analog circuit blocks realized by dint of the 90 nm complementary metal-oxide semiconductor technology are developed in this work for breast cancer analysis. The model's reliability findings are confirmed through comprehensive evaluation metrics, achieving sensitivity, specificity, accuracy, F1-Scoree, Intersection-over-Union, and Matthew's correlation coefficient of 0.9668, 0.9869, 0.9800, 0.9708, 0.9433, and 0.9556, respectively. Monte Carlo analysis further validates a 98.00 % accuracy on the Wisconsin Breast Cancer Dataset, with a power consumption of 23.06 mW, using 122 MB of memory while operating at a power supply rail of ±900 mV per analog circuit component and 27 °C computational unit temperature. These findings suggest a significant opportunity for integrating the proposed model into clinical practice, thereby improving patient care in medical diagnostics.