Researchers at the Kyushu University in Japan say they have developed a machine-learning model that can accurately evaluate the density of surviving cells in the malignant bone cancer osteosarcoma from pathological images which can provide a more accurate patient prognosis compared with conventional methods. The research is reported in njp Precision Oncology.
While surgery and chemotherapy have proven to be effective tools for the treatment of patients with localized osteosarcoma, those patients with metastatic disease have low rates of survival. Currently, after a patient has been treated with either surgery or chemotherapy, a necrosis rate assessment is performed, whereby a pathologist visually evaluates the proportion of dead tissue within a tumor. This assessment helps determine the ongoing treatment plan for a patient. Unfortunately, there is wide variability in the assessment of pathologists of the necrosis rate, which can lead to inaccurate prognoses.
With this understanding, the investigators at Kyushu University sought to develop a more nuanced assessment of the living versus dead tumor cells via the development of an AI-driven machine-learning model.
“In the traditional method, the necrosis rate is calculated as a necrotic area rather than individual cell counts, which is not sufficiently reproducible between assessors and does not adequately reflect the effects of anticancer drugs,” noted co-first author Makoto Endo, MD, PhD, a lecturer of Orthopedic Surgery at Kyushu University Hospital. “We therefore considered using AI to improve the estimation.”
To develop their model, the team first validated their method to detect surviving cancer cells using patient data, which showed it was capable of identifying viable tumor cells at the same level of proficiency as expert pathologists.
Once the model was validated, the investigators looked to analyze two key measures of osteosarcoma. First, they sought to determine disease-specific survival—the duration after diagnosis or treatment without death directly caused by the cancer. Second, they examined metastasis-free survival, which monitors the time after treatment without the cancer spreading to other parts of the body.
The researchers also determined the correlation between the AI model’s estimate of tumor cell density and prognosis and found it to have reproducible comparable detection and precision compared with a pathologist.
In the next step, the investigators sorted the patients into groups based on viable cell density either above or below 400 cells per square millimeter. Survival analysis of the two groups showed those in the high-density group showed worse prognosis than their low-density counterparts for both disease-specific survival and metastasis-free survival. Significantly, necrosis rate was not associated with either disease-specific survival or metastasis-free survival indicating that viable tumor cell density is a more reliable predictor of patient prognosis.
The researchers noted that the model’s measurement of viable tumor cells reflects its inherent malignancy and the individual tumor cell response in osteosarcoma. Using AI to analyze tumor pathology images can improve accuracy and eliminate the variability of human assessments. They also contend that the identification of viable tumor cells, which can continue to multiply after treatment, is a more reliable predictor of treatment response than cell necrosis.
“This new approach has the potential to enhance the accuracy of prognoses for osteosarcoma patients treated with chemotherapy,” Endo noted. “In the future, we intend to actively apply AI to rare diseases such as osteosarcoma, which have seen limited advancements in epidemiology, pathogenesis, and etiology. Despite the passage of decades, particularly in treatment strategies, substantial progress remains elusive. By putting AI to the problem, this might finally change.”
Next steps call for a large-scale validation of their machine-learning model to determine if it can be advanced for broader application in the clinic.
