This paper uses MRI images to introduce a novel machine-learning method that enhances deep learning algorithms for lung conditions diagnosis, including COVID-19. Therefore, the need for improved and faster, as well as easier-to-scale diagnostic methods mandated the start of the project in the healthcare field due to the challenges presented by MRI information. The model uses convolutional neural network (CNN) architecture optimized for high-dimensional MRI data, enabling the system to classify four major categories: COVID-19, Lung Opacity, and Normal & Viral Pneumonia. A relevant issue discussed in this work is the possibility of training models ourselves from scratch on some practical-scale, high-quality data set and the time-consuming process of building complex models. Considering these constraints, the advanced framework combines deep learning with the existing statistical analysis methodologies to improve or extend the model's stability. Applying them together with Machine learning and Deep learning, the predicted results are smoothed with logistic regressions to eliminate the overfitting issues commonly seen in deep learning-based models. For this purpose, the proposed model has incredible real-world accuracy and dependability to be important and efficient for medical applications. The results revealed test accuracy of 85.57%, while the categories’ precision, recall, and F1-score show orders of slightly better prediction. The framework provided is very flexible and must be designed to work with the medical datasets and environments specifically; therefore, its strength is its versatility. This hybrid model continuously improves classification accuracy and is much faster in time computation as it can be used in large, complex healthcare institutions. The study also demonstrates the feasibility of incorporating deep learning with conventional statistical techniques, thus giving a robust, feasible, and efficient solution for COVID-19 and other lung conditions using MRI data. The framework's flexibility when deployed in different environments indicates it can be useful for medical diagnosis, particularly in settings with high patient volume and the need for rapid decision-making.