human-machine cooperation with explainable AI and deep learning has a lot of potential to improve the diagnosis of cardiac problems in children. This strategy can result in more precise, transparent, and effective diagnoses, thereby improving patient treatment and outcomes, by fusing the knowledge of pediatric cardiologists with the strength of AI algorithms. To overcome the obstacles and guarantee that the AI system’s judgments are understandable, dependable, and in accordance with medical norms and laws, however, is crucial.

A promising strategy to increase the precision and reliability of diagnoses while simultaneously revealing insights into the thought process of the AI model is human-machine collaboration in the diagnostic process of heart disorders in pediatrics using explainable AI and deep learning. The strength of AI algorithms is combined with the knowledge of healthcare experts, notably pediatric cardiologists, in this partnership.


The appropriate assessment and interpretation of numerous clinical and diagnostic data are essential for correctly diagnosing heart disorders in young patients. By offering knowledge and assistance during the diagnosis process, explainable AI (XAI) and deep learning approaches can significantly help pediatric cardiologists. Using these technologies, the diagnostic procedure might proceed as follows:

Explainable AI (XAI)

Explainable AI refers to an AI system’s capacity to offer clear and understandable justifications for its choices. This is essential in medical applications as it aids doctors in comprehending the rationale behind the AI’s diagnostic recommendations. The system can increase confidence among medical professionals and guarantee the accuracy of its forecasts by offering explanations.

To ensure that the deep learning model’s judgments can be transparently comprehended by medical practitioners, combine explainable AI approaches with it. Attention processes, saliency maps, and feature visualization techniques are some examples of explainable AI strategies.

Deep Learning

A branch of machine learning that use multi-layered neural networks to automatically identify patterns and features from massive volumes of data. In medical image analysis, particularly the interpretation of echocardiograms, X-rays, and other medical imaging data required to diagnose cardiac disorders, deep learning has achieved exceptional success.

Create a deep learning model that can interpret input data and detect the presence of cardiac disorders, such as a convolutional neural network (CNN).

Heart disease in pediatrics

Pediatric congenital heart disease (CHD), commonly referred to as heart disease in children, is a broad term that refers to a variety of anatomical and functional cardiac defects that affect children from birth or emerge during childhood. It is a serious medical issue that affects about 1 in 100 neonates worldwide. Simple congenital malformations that may not require immediate medical attention to complicated congenital anomalies that necessitate emergency medical attention, these conditions can range in severity and complexity

Data Collection and Preprocessing

A comprehensive and well-annotated dataset of juvenile heart disease cases is needed in order to train the deep learning model. To guarantee the generalization and robustness of the model, this dataset should contain a wide range of situations, age groups, and demographic data. In order to clean, normalize, and enhance the data for efficient model training, data preparation techniques are used.

and there is a example of how to annotate vessel wall in HSA KIT software its in the shape of donut

zoom in view

zero capacity

this is full capacity

Model Training and Validation

Using the gathered and preprocessed data, the deep learning model is trained. To assess the effectiveness of the model, the dataset must be divided into training, validation, and testing sets. To increase the model’s precision and generalizability, strategies including cross-validation and data augmentation are employed.

this is a table of the names of files, number of pictures (ROIS) and blood vessel wall:

  Name of files  ROISNum of Blood vessel wall
  Rat104_Heart_HE_01 .ndpi  94  57
Rat105_Heart_HE_01 .ndpi3821