Reducing False-Positive Detections Using the Distance Between Activation Distributions in Individual Channels

We propose a novel method to estimate the confidence of outputted predictions of a convolutional neural network. We show that different channels in one layer can be treated as an ensemble and extract the confidence of a prediction from a single channel. To achieve this, we compute statistical distances between activation distributions located at the predicted mask and its surrounding area and aggregate it across all channels in a deep layer of a network. Research on a segmentation network of lung cancer nodules from 3d computer tomography images has shown growth of precision compared to the thresholding output network values. The more layers used to compute confidence, the better performance obtained, allowing for up to 18% fewer false-positives detections on the source Cancer dataset and up to 54% fewer false-positives detections on an unseen Covid dataset. Analyzing channel activations doesn't require any changes in the training procedure with a negligible amount of additional computations at the inference time. © 2022 IEEE.

A DEEP LEARNING BASED LOCATION AWARE ENCODING NETWORK ARCHITECTURE FOR LUNG CANCER SEGMENTATION

Lung cancer (LC) is one of the most common malignant tumors, with rapid growth and early spread. LC is one of the most common malignant tumors. Lung cancer is a deadly disease, and early detection is essential. To achieve more precise diagnoses, cancer segmentation aids clinicians in determining the extent and location of cancer. But manually segmenting lung tumors from large medical images is a time-consuming and difficult task. A convolutional neural network (CNN)-based encoding network with position awareness is proposed in this study for automatically segmenting LC from computed tomography images. It is our model's design philosophy to change the usual link net architecture so that we can properly identify cancer. Our innovation resides in the manner we connect each encoder with decoder, in contrast to previous neural network topologies used for segmenting. During the encoder's many downsampling processes, spatial information is lost. By employing simply the encoder's down sampled output, it is impossible to retrieve this lost information Through the use of untrainable indices, the encoder and decoder are connected together. The output of an encoder may also be sent straight into a decoder, which can then execute segmentation on it.To conduct this study, a spatial attention-based encoder and a decoder that bypasses each encoder's input to the output of its related decoder were employed. Decoding and upsampling procedures will benefit from the spatial information that is recovered in this manner. With each layer of encoded information, the decoding process may require less parameter space, making it more efficient. Lung Image Database Consortium image collecting dataset obtained 98.5 percent accuracy in verifying the suggested system's performance. According to the study mentioned, a subjective comparison between the suggested approach and certain current methodologies is also carried out. Experiments have shown that the suggested method outperformed current technologies, allowing radiologist to more precisely locate a lung tumour while using it.