Background/problem
Colorectal cancer is the fourth leading cause of death and accounts for 15.6% of total deaths linked with cancer worldwide. Late detection dramatically reduces its five-year survival rate and increases treatment costs. Polyps (precancerous or cancerous growths) develop in the early stages of this cancer, and an important goal of colorectal cancer screening is to detect and remove these polyps. Colonoscopy is the current gold standard, and yet it is still flawed as it is heavily physician-based and prone to human error. Current methods of colonoscopies can result in up to 25% missed polyps and misclassification. These false results may lead to unnecessary hospital or physicians’ visits and increased time, costs, and risks for patients.
Artificial intelligence (AI) is anticipated to enhance the accuracy of screening tests and aid endoscopists in identifying lesions during colonoscopy. Different models, ranging from simpler algorithms like SVM and KNNs to more intricate deep neural networks, have been employed for this purpose. However, these interventions have often been challenged by limited and/or biased datasets that lead to overfitting issues, undermining effectiveness, reliability, and accuracy in clinical settings.
Tech overview/solution
UT Austin researchers present a novel suite of AI-algorithms for the diagnostic imaging process. The patent-pending technology uses a holistic approach, comprising five modules, each of which contributes to the transparency, user-input friendliness, and reliability of clinician-AI interactions.
- The Cascade Reliability Framework (CRF) generates a collection of “potential classes” as predicted by the base algorithm, accompanied by accurate probability estimates for each class.
- The Vision Transformer (ViT) provides a set of possible labels and their associated confidences and highlights the region of interest that informed the model’s decision.
- The Explainer adds additional context to the model predictions (CRF + ViT) using descriptive text.
- The GenSynth generates synesthetic medical images based on “seed images” paired with text prompts.
- The TeachIt module allows clinicians to cross-evaluate existing images, input their own images and descriptions, and further train the model.
Benefits/competitive advantage
The technology addresses a high unmet need of imaging and diagnostic markets where AI-assisted colonoscopies will accelerate diagnoses, more accurately and precisely detect lesions, and recognize pathology. With the modular framework of the suite, key providers can “take one” or “take all,” depending on their need to provide creative, effective, and optimal diagnostic solutions for their clinicians and patients. This approach to AI-assisted diagnosis also ensures that clinicians benefit from the latest advancements in deep learning technology, while maintaining full transparency and control over the decision-making process.
Opportunity
With increasing prevalence rates, increasing geriatric population, and prevention/early detection awareness and screening programs, advanced imaging diagnostics for colorectal cancer represent a significant market opportunity for companies over the next five years. Global colorectal cancer diagnostics industry is predicted to expand at a CAGR of 8.5% with a projected market value of $4.5B in 2031 (a 2.2× rise from 2020). This technology is a great fit for companies with an interest in: AI-powered imaging, advanced imaging, diagnostic solutions, colorectal cancer, colorectal cancer screen, colonoscopy, and rapid screening.
