The healthcare vertical today is patient centric and data driven with the advances in IOT and Artificial Intelligence. A need for early detection and diagnosis for any contiguous diseases or infections is required which is generally performed through radiographic analysis. Deep learning in the field of radiologic image processing reduces false-positive and negative errors in the detection and diagnosis of disease and offers a unique opportunity to provide fast, cheap, and safe diagnostic services to patients.
Deep learning has the potential to augment the use of chest radiography in clinical radiology, but challenges include poor generalizability, spectrum bias , and difficulty comparing across studies. Recently, several clinical applications of CNNs have been proposed and studied in radiology for classification, detection, and segmentation tasks. The Deep learning deterministic model predicts the radiograph into three classes such as Normal, Covid and Viral Pneumonia. The probabilistic model predicts the radiograph into Normal, Cardiomegaly, mass and other abnormalities using Class activation maps(CAM). The main objective is to develop a deep learning–based reconfigurable architecture that can classify normal and abnormal results from chest radiographs with major thoracic diseases including pulmonary malignant neoplasm, active tuberculosis, pneumonia, pneumo-thorax, covid-19 etc, and to validate the performance on Intel FPGA.

Project Proposal

1. High-level project introduction and performance expectation

Deep learning has cutting-edge technology and has application in every field of life ranging from computational to healthcare. It has a very deep impact on the life of the people or societies because its application is always the need of the day. Radiography is an imaging technique using X-rays, gamma rays, or similar ionizing radiation and non-ionizing radiation to view the internal form of an object. Applications of radiography include medical radiography (”diagnostic” and ”therapeutic”) and industrial radiography.  Medical radiography can be broadly classified as RAT (Rapid-Antigen test) , RT-PCR (Reverse transcription polymerase chain reaction) , CXR (Chest X[1]RAY) and lastly CT-SCAN (computed tomography). Chest imaging is a quick and easy procedure recommended by medical and health protocols and has been mentioned in several texts as the first tool in screening during epidemics. Chest radiography is convenient and fast for medical triaging of patients. Unlike CT scans, X-ray imaging requires less scarce and expensive equipment, so significant savings can be made in the running costs. Furthermore, portable CXR devices can be used in isolated rooms to reduce the risk of infection resulting from the use of these devices in hospitals. The main objective is to develop a deep learning–based reconfigurable architecture that can classify normal and abnormal results from chest radiographs with major thoracic diseases including pulmonary malignant neoplasm, active tuberculosis, pneumonia,  pneumo-thorax, covid-19 etc, and to validate the performance on Intel FPGA.

2. Block Diagram

3. Expected sustainability results, projected resource savings

The different use cases such as language translation, game and decision systems, medical diagnosis, social media applications, robotics, advanced driver assistance systems and deep embedded systems requires different computational networks & different figures of merits: speed, latency, energy, accuracy which in turn requires flexibility in a deep learning solution. Most of the deep learning algorithms require billions of MAC operations and large amounts of data to store model parameters. This requires smart architectural choices of how tightly to couple memory and compute. This has been clear in the last few years that researchers are actively looking for models which will provide higher accuracy, or integrating various models to solve new types of problems which requires flexibility of implementation. With growing maturity of the deep learning field, we are seeing divergence in need for training and inference. Training is focused of developing models which will improve accuracy of a deep learning task while reducing training time. There is an increased emphasis on using trained models to predict/estimate outcomes from new observations in efficient deployments. FPGAs provide custom hardware levels of performance at the same time provide flexibility to design new circuits as application needs change. Inference is the process of running a trained neural network to process new inputs and make predictions. Training is usually performed offline in a data center or a server farm. Inference can be performed in a variety of environments depending on the use case. Intel® FPGAs provide a low power, high throughput solution for running inference.