Explainable AI for Endometriosis Diagnosis: A Dense Neural Network Approach with SHAP Interpretation

According to [2], Artificial Intelligence’s revolutionary potential is bringing about a fundamental revolution in every aspect of the world, impacting everything from product development to medical diagnosis. It also has several subfields, particularly in the practice of medicine, including computer vision (CV), deep learning (DL), and machine learning (ML). Women’s quality of life is greatly impacted by endometriosis, a chronic gynecological disorder that can lead to infertility and discomfort [10]. Even though it is common, non-invasive diagnosis is still difficult. Researchers have used information from imaging, blood tests, genetics, and symptoms to investigate different Machine Learning (ML) methods. Methods such as LASSO regression and logistic regression have demonstrated potential [11]. Significant restrictions still exist, nonetheless, which impair patient outcomes and clinical adoption. First, current research frequently relies on a small number of data sources or a single data type, which may cause important information that could improve diagnostic accuracy to be missed. For instance, research that only looks at symptom data may overlook important information from other sources. Second, many machine learning models used in contemporary research are “black boxes,†which means that their decision-making procedures are opaque [12]. This lack of interpretability limits trust and prevents broad clinical adoption by making it challenging for medical professionals to comprehend the process used to generate diagnosis. When using XAI to diagnose endometriosis, decision confidence and trustworthiness are greatly increased (Antoniadi et al., 2021). By having a thorough understanding of how an AI system makes a diagnosis, clinicians can make well-informed decisions regarding its application, which will increase the degree of trust and adaptability of the technology in the medical field. Based on the features it prioritizes, it can detect any bias in the model’s predictions.