Data acquisition is often an overlooked aspect in the medical sector. With the rapid advancement in machine learning and artificial intelligence, the need for clean and structured data has significantly increased. There have been many advances in this field through LLM techniques and Generative AI with the advent of models like Llama. However, these technologies have crucial limitations. These models have poor portability, making them difficult to use in a resource-constrained environment. They often require high-end computing utilities like GPUs or extensive API calls, both of which act as bottlenecks and may introduce significant latency. To address this issue, we propose a technique with a Convolutional Neural Network (CNN) and advanced image processing algorithms to detect data frames, namely tables, images, and graphs, which are the key data components in a file. The graphs and images are singled out for further processing. Meanwhile, the tables are converted into data frames, which are more flexible for processing as they are easily convertible to other formats and are lightweight. The approach enhances scalability by eliminating the need for multiple API calls, which reduces dependency on external services, minimizes latency, and enables faster, more efficient data processing in real-time scenarios. Experimental results show that our model achieves detection accuracy comparable to high-end models while operating significantly faster. For graph and image detection, median-based filtering demonstrated superior accuracy (92%) and a lower false positive rate, where as mean-based filtering was faster and more suited for structured documents. The model also generalizes well across datasets, maintaining strong performance despite being trained on a single dataset. This solution provides an efficient, portable alternative for medical data acquisition and processing, offering accuracy and speed while remaining adaptable for future extensions.