Deep Learning Analysis of Chest Radiographs for Differentiating Between Congenital Ventricular Septal Defect and Atrial Septal Defect in Children

Children with atrial septal defect (ASD) and ventricular septal defect (VSD) are commonly examined for respiratory symptoms, even when the underlying condition is not identified. Radiographs of the chest are frequently used as the main imaging technique. As ASD and VSD are treated differently, it is important to distinguish between the two. The study investigated the effectiveness of deep learning analysis of chest radiographs in differentiating between ASD and VSD in children.

A total of 1,194 patients' chest radiographs and radiology reports were examined, with cases divided into training, validation, and test sets. Four deep learning models including ResNet-CBAM, InceptionV3, EfficientNet, and ViT were developed and optimized using fivefold cross-validation, and their performance was evaluated through receiver operating characteristic (ROC) curve analysis.

The results demonstrated the following findings:

• All models achieved over 70% accuracy, sensitivity, and specificity in the differential diagnosis of VSD and ASD.
• InceptionV3 showed the highest performance at 87% classification accuracy, surpassing the accuracy of two radiologists. 
• The calculated the area under the ROC curve (AUC) values for the models were 0.87 (ResNet-CBAM), 0.91 (InceptionV3), 0.90 (EfficientNet), and 0.66 (ViT).

In conclusion, deep learning techniques, particularly InceptionV3, based on chest radiography, showed promising capabilities in accurately distinguishing between VSD and ASD, potentially helping radiologists in diagnosis, education, and reducing errors in clinical practice.

Source: Jia H, Tang S, Guo W, Pan P, Qian Y, Hu D, Dai Y, Yang Y, Geng C, Lv H. Differential diagnosis of congenital ventricular septal defect and atrial septal defect in children using deep learning-based analysis of chest radiographs. BMC Pediatr. 2024 Oct 15;24(1):661. doi: 10.1186/s12887-024-05141-y. PMID: 39407181; PMCID: PMC11476512.