The Role of Weight Initialization and Preprocessing Techniques in Analyzing Chest X-ray Images with Deep Neural Networks: A Comparative Study
Year: 2026
Type: article
Abstract: Accurate interpretation of chest X-rays (CXRs) using machine learning techniques plays a vital role in improving diagnostic practices within healthcare. This study advances CXR analysis by exploring various weight initialization and preprocessing methods with deep neural networks. Specifically, (i) an object detection model was trained to identify regions of interest (ROIs) within CXRs, enabling focused image analysis; (ii) weight initialization strategies, including partial fine-tuning, complete fine-tuning, and random initialization, were assessed using the EfficientNet-B1 model on both original and cropped images; and (iii) preprocessing techniques such as histogram equalization, bilateral filtering, Gaussian filtering, and Contrast Limited Adaptive Histogram Equalization (CLAHE) were applied to enhance image quality. Results show that ROI selection significantly improves model performance by focusing on relevant image areas (+4.8%). Pre-trained ImageNet weights with complete fine-tuning outperformed random initialization (+6.5%), demonstrating the advantages of transfer learning, particularly with limited datasets. Among preprocessing methods, the combination of CLAHE with Gaussian filtering achieved the highest validation accuracy (0.958), suggesting that advanced preprocessing methods substantially enhance model performance. These results underscore the importance of effective weight initialization and preprocessing in optimizing diagnostic accuracy and efficiency in CXR analysis.
Source: DÜMF Mühendislik Dergisi
Author: Ömer Faruk Söylemez
Institution: Dicle University
Language: Turkish
Topic: COVID-19 diagnosis using AI
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7140312846