Cancer Misdiagnosis – a Challenge for Pathologists
Healthcare systems around the world, regardless of whether they operate in a public, private or some kind of a hybrid model, all strive to provide quality healthcare. At the same time, cancer incidence is on the rise worldwide, resulting in a growing number of biopsy tests. This, in addition to a global shortage in pathologists, creates challenges for diagnostic laboratories, that need to analyze more tests while maintaining their high quality standards.
Accurate diagnosis by a pathologist is usually the first step in the cancer pathway. Misdiagnosis at this point will severely impact treatment decisions, patient survival rates and ultimately their mental well-being. This is most evident in the case of false-negative type errors, where a patient with cancer is advised that his biopsy was benign, leading to delayed detection and treatment, sometimes with fatal consequences. But how many misdiagnosed cancers are there? Estimates on error rates in pathology range between 3-5%, while in reality this number is considered a gross underestimation, since misdiagnosis is difficult to detect and rarely tracked, documented, or made public. What is well recognized is that pathologists are severely overworked, many labs are understaffed and the majority of cancer diagnosis is still conducted by looking at biopsies through a microscope, a 100 year-old practice.
The increasing adoption of new technologies in pathology, more specifically digital pathology and slide digitization, as well as advancements in artificial intelligence (AI) and machine learning, have opened possibilities for developing new solutions that provide pathologists with decision-support and quality control tools. One such solution is Ibex’s Galen™ Prostate, a CE-Marked solution which includes a Second Read application that analyzes prostate biopsy cases using an AI powered algorithm, subsequent to the pathologist’s diagnosis. The application provides pathologists with an immediate comparison between their diagnosis and the algorithm’s findings and supports them by alerting in case of discrepancies with high clinical importance (e.g. a missed cancer). Using such a safety net, pathologists can reduce diagnostic errors in the lab by enhancing quality control.
Can an AI Algorithm Detect Missed Cancers?
With an AI-powered solution available for trial, the immediate question for a pathologist would be: Does it really work in the field, in a blinded and independent environment? in order to provide robust scientific data to address this question, Ibex partnered with Medipath, the largest network of pathology labs in France. The two companies announced a strategic partnership in 2019 and went on with a clinical study looking at the performance of the Ibex algorithm in diagnosing prostate cancer - the most frequently diagnosed type of cancer in men, and evaluated the potential of using an AI-based solution for quality control.
The Ibex prostate algorithm is designed to help pathologists detect prostate cancer and other clinically significant features such as the Gleason grade, perineural invasion, high grade PIN and more. The algorithm was trained on slides from a dataset of over 60,000 prostate slides from multiple institutions and representing a variety of diagnoses and clinical findings.
The study included 100 patients who underwent prostate biopsy, all originally diagnosed as not having cancer (the study set included a total of 801 slides). Ibex’s prostate algorithm analyzed each of the slides and identified 25 slides from 15 different patients as including features suspicious of cancer. These suspicious slides were sent for review by three blinded independent pathologists. With respect to 15 slides from 12 different patients, at least one reviewer determined that they were cancerous. Compared with the results of the original benign diagnosis, this represents a misdiagnosis rate of 12%. Interestingly and unexpectedly, there were misdiagnosed cases with a relatively high Gleason score (e.g. 4+4 or 3+5) and tumor size larger than 3mm.
Conclusions
Missed prostate cancer, the most common cancer in men, is not a rare event. In fact, contrary to commonly held conceptions in the field, misdiagnosed prostate cancers are not limited to low-grade or small size tumors and higher-grade and larger size tumors are also missed. The Ibex algorithm was able to correctly identify these missed cancers, and if used in a live setting would have prevented their misdiagnosis. In view of these results, pathology institutes could benefit from using field-proven and accurate AI-based solutions such as the Galen™ Prostate, that was recently CE-Marked, for analyzing all the biopsies in parallel to the pathologists’ reviews, helping them improve quality control and reduce diagnostic error rates.
Examples of Missed Cancers and Other Features Detected by The Ibex Algorithm
The study was presented at the 2019 European Congress of Pathology (Read the full poster)
ABOUT IBEX
We are a multidisciplinary team of entrepreneurs, data scientists, software engineers and medical experts, working together to realize our vision: Applying AI to cancer diagnostics, striving to improve patient outcomes and quality of life.
