MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641074914 A) filed by Sasi Institute Of Technology & Engineering; A. Durga Bhavani; A N V K Swarupa; I. Usha; and K. Sandeep on June 17, 2026, for Hybrid Cnn–transformer Model For Automated Lung Cancer Detection Using Ct Scan Images.

Inventors include Sasi Institute Of Technology & Engineering; A. Durga Bhavani; A N V K Swarupa; I. Usha; and K. Sandeep.

The application for the patent was published on June 26, 2026, under issue no. 26/2026.

Abstract: Lung cancer is a leading cause of cancer-related mortality worldwide, with late diagnosis significantly affecting patient survival. Although CT imaging is the standard method for pulmonary screening, manual interpretation is time-consuming, subject to variability, and may overlook small nodules. These limitations create a need for automated computer-aided detection (CAD) systems. This project presents a deep learning-based framework for automated detection and classification of pulmonary nodules from CT images. The system utilizes the LUNA16 dataset for normal CT scans and the U5000 DICOM dataset for cancer-positive cases. The preprocessing pipeline includes image loading, min-max normalization, quality filtering, dataset balancing, resizing to 224 × 224 pixels, data augmentation, and dataset preparation. The method operates directly on filtered CT slices without segmentation masks. A hybrid EfficientNet–Vision Transformer (ViT) model is proposed for binary classification. EfficientNet extracts local features, while ViT captures global contextual information. The extracted features are fused and processed through fully connected layers with Batch Normalization, ReLU activation, and Dropout before generating the final classification output. The model is trained using BCEWithLogitsLoss and the AdamW optimizer with augmentation techniques such as Gaussian blur, Gaussian noise, and brightness adjustment. During inference, Test-Time Augmentation (TTA) with flip and rotation transformations is applied, and prediction averaging is used to improve robustness and classification performance.

Disclaimer: Curated by HT Syndication.