Automated Detection of Malignant Lesions in the Ovary Using Deep Learning Models and XAI

AI models can achieve high accuracy in detecting ovarian cancer: The study found that an InceptionV3 deep learning model achieved 94% accuracy across multiple performance metrics in identifying malignant ovarian lesions, suggesting AI could help address the current diagnostic gap where non-invasive detection methods are unreliable.

Explainability is essential for clinical adoption: The researchers used three different explainable AI (XAI) techniques—LIME, Integrated Gradients, and SHAP—to make the AI model's decisions transparent, recognizing that doctors need to understand why the AI made a diagnosis before trusting it in real clinical settings.

Multiple AI approaches were tested to find the best solution: Rather than relying on a single model, the team developed 15 different variants using various neural network architectures (LeNet-5, ResNet, VGGNet, and Inception), demonstrating that responsible AI development requires systematic comparison to identify the most reliable approach.

AI could reduce the burden of invasive procedures: By improving non-invasive detection accuracy, this technology addresses a real clinical problem—ovarian cancer currently requires time-consuming and invasive procedures for accurate diagnosis—potentially sparing patients from unnecessary procedures.

Dataset quality matters for responsible AI: The study used a specific histopathology dataset and employed data augmentation techniques, highlighting that responsible AI systems depend on properly curated medical data to ensure reliable and generalizable results.