An IoT-Enabled Deep Learning Approach Implemented on Android Device for Automated Identification of Breast Cancer Using Thermal Images

Breast cancer (BC) is a very common type of cancer in women, and it occurs due to the abnormal growth of breast cells to produce malignant tumors. The early detection of BC is challenging in clinical standards to reduce the fatality rate caused by this disease. Artificial intelligence is helpful in early and automated detection of BC and provides a cost-effective way to assist radiologists in providing better diagnostic decisions. The artificial intelligence (AI) model integrated with the Internet of Things (IoT) provides the framework for real time analysis of patient data and tele-healthcare monitoring for detecting BC. This paper proposes a novel IoT-enabled deep learning based approach implemented on an Android device to detect BC using thermal images. A deep convolutional neural network (CNN) architecture with five blocks of cascaded convolutions followed by max-pooling after each block and cascaded dense layers is formulated and trained using the Google Cloud central processing unit. The post-training quantization (PTQ) of deep CNN (DPCNN) is performed using floating-point 16-bit (FP16) and integer 8-bit (INT 8)-based quantization techniques. The reduced-size DPCNN model is deployed on a cloud framework and an Android device for real-time detection of BC using thermal images. The DPCNN model deployed on the Android device provides a portable framework for low latency, enhanced privacy, and offline processing compared to the cloud-based framework for detecting BC using thermal images. The experimental results obtained using a public database reveal that the proposed DPCNN has obtained the accuracy values of 99.63% and 99.27% for FP16 and INT8 cases to detect BC. The proposed DPCNN model has fewer parameters and higher classification performance than transfer learning and existing methods in detecting BC using thermal images.