CT - Based Organ Anomaly Detection Using U-Net and Convolutional Neural Networks Hybrid Technique

Abstract

Deep learning-based medical image analysis advanced significantly with transformer-based architectures and self-configuring segmentation frameworks. This paper presents an implementation-focused approach for CT-based organ anomaly detection using deep learning techniques. The proposed system integrates the U-Net framework for accurate and automated organ segmentation, ensuring adaptive performance across different CT datasets without manual tuning. For feature extraction, convolutional neural networks (CNNs) and transformer-based models are employed to capture both local spatial patterns and global contextual relationships in CT images. Anomaly detection is performed using classification-based methods to distinguish normal and abnormal regions, along with reconstruction-based techniques such as autoencoders to identify deviations through reconstruction error. The integration of these methods improves robustness and accuracy in detecting organ abnormalities. The primary objective of the system is to enhance diagnostic precision, reduce radiologist workload, and enable automated identification of abnormal regions in CT scans, contributing to efficient computer-aided diagnosis systems in healthcare applications. Accurate detection of organ anomalies from Computed Tomography (CT) scans is critical for early diagnosis and treatment planning. This paper presents a hybrid deep learning framework combining U-Net for organ segmentation and Convolutional Neural Networks (CNNs) for anomaly classification. The proposed approach first segments the target organ using a U-Net architecture and subsequently classifies the segmented region into normal or abnormal categories using a CNN model. Experimental results demonstrate improved accuracy and robustness compared to standalone classification models.