Seminars

View all Seminars  |  Download ICal for this event

Design of AI-Based Computational Framework for Accurate Detection of Polycystic Ovarian Disease and Ovarian Cancer Using Ultrasound, CT, and Histopathology Images

Series: Ph.D. Colloquium

Speaker: Ashwini Kodipalli, Ph.D (Engg.) ERP student, Dept. of CSA

Date/Time: Jan 13 11:00:00

Location: CSA Auditorium, (Room No. 104, Ground Floor)

Faculty Advisor: Dr. Susheela V. Devi (Retd.) and Prof. Chiranjib Bhattacharyya

Abstract:
Polycystic Ovarian Disease (PCOD) and ovarian cancer represent critical health challenges affecting millions of women globally. Early and accurate detection of these conditions is essential for effective clinical management and improved patient outcomes. However, the diagnostic process is often hindered by inter-observer variability, limited access to expert clinicians, and the complexity of analyzing high-dimensional medical imaging data.

This research proposes a comprehensive, computer-aided computational framework leveraging deep learning techniques to address these challenges. The framework focuses on accurately diagnosing PCOD using ultrasound images and ovarian cancer using computed tomography (CT) scans, histopathology images, and clinical data. A key aspect of this work is the emphasis on real-world clinical datasets, which present unique challenges such as inconsistencies, noise, and variations in image quality. Robust preprocessing methods are developed to standardize and prepare the data for analysis.

For PCOD diagnosis, Variational Autoencoders (VAEs) are employed to augment the dataset, addressing the issue of limited data availability. Semantic segmentation and classification are performed using UNet and Attention-based UNet models to extract informative regions from ultrasound images. The Attention-based UNet performs better, achieving higher Jaccard and Dice scores than the vanilla UNet. The segmented regions and corresponding labels are then used to train multiple weak learners, including random forests, support vector machines, and Efficient Net models, to map features to diagnosis labels. A novel approach combines predictions from weak learners with segmented data to construct appended feature sets. These stacked features are utilized to train an artificial neural network, achieving an impressive classification accuracy of 98.12% on the test set.

For ovarian cancer detection using CT scan images, a novel image-to-image translation approach is proposed to address the challenge of prolonged training and inference times caused by the large size of CT images. A UNet model is trained with expert-annotated boundaries to generate segmentation patches for benign and malignant tumors, achieving Intersection over Union (IoU) scores of 0.820 and 0.775, respectively. A conditional GAN is employed to generate artificial segmentation patches similar to the ground truth, with IoU scores of 0.825 (benign) and 0.765 (malignant) and per-pixel class accuracies of 80.05% (benign) and 72.70% (malignant). These patches are used to train a ResNet-101 model for tumor detection, achieving accuracies of 82.5% (benign) and 80.0% (malignant). Additionally, an enhanced ResNet-50 architecture is proposed for CT scan classification, achieving an accuracy of 97.5%.

For ovarian cancer detection using histopathology images and clinical data, a 3-stage ensemble model was developed for classifying ovarian cancer using clinical data, achieving an accuracy of 98.13% with random forests. The support vector machine demonstrated the highest generalizability among conventional learners, achieving an accuracy of 97.06%. The predictions were explained using XAI methods, including LIME (local interpretability) and SHAP (global interpretability). Statistical validation of model performance was conducted using p-tests and Cohens d-tests. Additionally, a novel ResNet-56 architecture was proposed for classifying histopathology images, achieving an accuracy of 98.69% on the training set and

97.3% on the test set, outperforming models such as VGG16, VGG19, InceptionV3, ResNetV2, and Xception.

The final objective integrates clinical data and histopathological images into a single pipeline using contractive autoencoders to generate embeddings that capture the rich features of both modalities. These embeddings are concatenated to form a fused feature vector used to train a novel Inception ResNet-118 network. The proposed network achieves a classification accuracy of 98.82%, significantly outperforming contemporary models like ResNet-50, DenseNet201, and MobileNet. This result underscores the effectiveness of combining complementary information from multiple modalities for tumor diagnosis.

The application of this research extends beyond mere detection to providing reliable, scalable, and explainable diagnostic tools for clinical use. By addressing key challenges in medical imaging, this work enables the early diagnosis of PCOD and ovarian cancer, which is critical for timely intervention and improved survival rates. Furthermore, the integration of deep learning models with real-world clinical datasets offers a pathway to making healthcare more accessible, enabling advanced diagnostic capabilities even in resource-limited settings. Ultimately, this research contributes to the well-being of mankind by enhancing diagnostic accuracy, reducing healthcare disparities, and improving patient outcomes on a global scale.