An Integrated Deep Learning Platform to Optimize and Forecast Surgical Supply Demand
According to McKinsey, the average supply chain has a digitization level of merely 43%, with an annual spending of $25.7 billion more on supply chain products than necessary. The COVID-19 pandemic, wildfires, and a massive explosion led to the most drastic disruptions of the global supply chain with hospital shortages, inexperienced surgeons in low-income hospitals, manual inventory management, and a lack of visibility. Now, more than ever, a digital transformation with integrated platforms and automation is required. Currently, there is no innovative technology that allocates resources based on images and integrates computer vision into the modern supply chain.
The goal of this project is to develop an integrated platform that diagnoses a patient’s immediate condition through computer vision and medical imaging, provides a surgical supply inventory based on the patient’s diagnosis, forecasts supplies through a thorough trend-based algorithm, and may be run on a smartphone.
To achieve the maximum automation, I divided execution into 4 phases: diagnose, allocate, forecast, and integrate. I developed a convolutional neural network through the DenseNet 121 architecture with generalizable features that can detect the type of melanoma present. The HAM10000 dataset was perfect in achieving this purpose, as it contained 10,015 dermoscopic images fully labeled by clinicians and allowed the model to be applicable to other skin lacerations typical of an emergency. Each feature (weights, pre-training, data augmentation, and early stopping) was eliminated to retrieve the most accurate and generalizable model, reaching an accuracy of 82%. In order to obtain the surgical tools or techniques to treat the diagnosed condition through the computer vision model, a resource allocation algorithm combined with a recommendation engine will be developed. Further steps include forecasting demand based on a time-series analysis of patient hospitalizations and an integration of the models into a smartphone through CoreML.
With all these features, the final product, MedBay, is the first to tie computer vision and supply chain economics together. The novel computational pipeline can be understood by anyone, regardless of an AI background, and applicable to any dataset. In addition to significant economic ramifications on medical and supply chain industries, MedBay would increase visibility, education, and quality of care. This multifaceted platform hopes to bridge the major gaps in our healthcare supply chain.