A Deep Learning Approach for Breast Tumour Diagnosis & Classification from Fine-Tuned EfficientNetB3 Model
Rudresh Pillai, Neha Vaishnavi Sharma, Rahul Singh Chauhan, Devyani Rawat, Rupesh Gupta · 2023
This study uses cutting-edge digital image processing techniques to diagnose and categorize breast tumours. By using the EfficientNetB3 transfer learning model and including fine-tuning layers, this research aims to improve the precision and effectiveness of breast tumour diagnosis. The dataset utilized in this research consists of 1134 breast ultrasound pictures, classified into three categories: normal breast tissue, benign tumours, and malignant tumours. The dataset has been carefully partitioned into training, validation, and test subgroups to assist the model's creation and assessment. The pre-existing EfficientNetB3 model is adapted and optimized through transfer learning to specifically cater to analyzing breast ultrasound pictures. The use of fine-tuning layers allows for the adaptation of the model to the distinctive attributes associated with breast tumour diagnosis. Throughout the training phase, the examination of accuracy and loss graphs offers valuable insights into the learning process of the model. This analysis facilitates the prompt determination of any possible difficulties that may arise. The research culminates with assessing the predictive model's efficacy on the test subgroup. A constructed confusion matrix provides a complete evaluative instrument, facilitating the computation of measures such as accuracy. The subsequent stage demonstrates the model's capabilities, giving insight into its possible implications for treatment. The initial results suggest that the proposed technique has shown an accuracy of 88%, highlighting its potential. Although this study shows potential, it also highlights the need for more research in several areas, such as dataset expansion, exploration of other architectures, and incorporation of clinical perspectives. Integrating advanced technology and medical diagnostics presents novel opportunities for enhancing breast tumour diagnosis. The fine-tuned EfficientNetB3 transfer learning model has shown promise as a vital asset in the battle against breast tumours.