FPGA Implementation of CNN for Lung Cancer Diagnosis

Abstract

Lung cancer is a fatal disease taking more than 1.8 million lives every year, necessitating timely and accurate diagnosis for effective treatment. This research investigates the utility of deep learning models to diagnose lung cancer and its hardware implementation on an FPGA (Field Programmable Gate Array). Notably, this research is distinguished due to the utilization of a dataset consisting of bone scans. The dataset comprises bone scans of more than 3247 patients, where some cases exhibit bone metastasis. This dataset undergoes stages of comprehensive processing to standardize image resolutions and remove any potential artifacts. Subsequently, CNN (Convolutional Neural Network) models are trained and evaluated using these bone scans in order to extract relevant features and classify them according to the presence or absence of metastatic lung cancer. The performance of the four CNN architectures and hyperparameter configurations is evaluated using accuracy, precision, recall, and F1 score metrics. A hardware implementation of the trained model is realized on an FPGA due to its efficient and parallel processing capability, enabling effective diagnosis. In the latter chapters, this thesis evaluates the performance and computational efficiency of FPGA, considering factors such as resource utilization, inference speed, and power consumption.