Novel AI-Based Web Tool Shows Promise for Prostate Cancer Grading, Bringing AI A Step Closer to the Clinic

Digital Pathology

by Christos Evangelou, MSc, PhD – Medical Writer and Editor


In a new study, researchers at Northeastern University, Maine Health Institute for Research, and Maine Medical Center developed a web application that combines human expertise with AI-driven grading predictions. The application incorporates diverse data, such as electronic health records, biopsy images, and report summaries.1

Feedback from medical professionals provided critical input on the utility of the tool. Overall, feedback on usability was positive, demonstrating the platform’s potential for integration into prostate cancer diagnosis. Suggestions for improvement included alternative chart formats and higher-resolution biopsy images.1

“This work will provide a more accurate and timelier diagnosis and grading than the manual existing process in prostate cancer diagnosis,” said Saeed Amal, a bioengineering professor at the Roux Institute at Northeastern who led the development of the web tool. “This will reduce the time for diagnosis, grading, and treatment, and will result in greatly improved patient care and help alleviate the issues caused by the nationwide shortage of pathologists.”

The report was published in Cancers (Basel).


Rationale: Aligning AI With Existing Physician Workflows

Prostate cancer continues to be a leading cause of cancer-related deaths among men in the United States.1 Accurate grading of prostate cancer is pivotal for determining its severity and guiding treatment decisions. However, integrating advanced AI techniques into real-world clinical diagnosis remains an ongoing challenge. This research tackled two key limitations of previous studies: narrow datasets and simulated environments that fall short of real diagnostic complexity.

To address these limitations, researchers have developed an interactive platform that aligns AI with real-world clinical needs, incorporates diverse data modalities beyond images, and mimics real-world diagnosis.

“Overcoming challenges such as a shortage of pathologists and grading variability in diagnoses is essential. Recognizing the potential of AI algorithms in real-world clinical settings, we developed a web application,” Prof. Amal explained.


Approach: Combining Human Expertise, Multimodal Data, and AI Prediction

The web application was designed to be compatible with clinical workflows for real-world diagnosis. Users can securely upload digitized prostate biopsy images and access AI-generated International Society of Urological Pathology (ISUP) grades, interactive visualizations highlighting regions of interest, electronic health records, and natural language processing (NLP) report summaries.

“This design ensures an efficient and meaningful interactive experience for users,” said Prof. Amal.

Upon processing biopsies, pathologists access AI-generated probability distributions over different Gleason cancer severity scores, annotated visualizations highlighting suspicious regions for inspection, and interactive tabs covering diverse analytical facets — from patient demographics to consolidated summaries. This expansive feature set provides multifaceted support for diagnostic decisions.


Evaluating User Perspectives

To assess the utility of the web application, researchers gathered perspectives from four pathologists and one medical practitioner. The survey participants used the application’s array of tools to evaluate prostate biopsies and provided their feedback via a 7–12-minute survey.

“The goal was to gather feedback from pathologists who used the application, enabling continuous improvement and refinement of the system for enhanced clinical efficacy,” noted Prof. Amal.

The questionnaire consisted of over 20 descriptive, rating-based, and multiple-choice questions evaluating key usability aspects, including ease of understanding and navigation, quality of data presentation and visualizations, perceived acceptability, and willingness for adoption. Participants also completed the validated NASA Task Load Index (TLX), which quantitatively assesses workload across six independent subscales: performance, effort, user frustration, and mental, physical, and temporal demands.

Open-ended feedback allowed participants to provide detailed suggestions, concerns, and impressions. This comprehensive mixed-methods approach aimed to illuminate strengths and limitations from multiple angles.


Key Findings: High Usability and Acceptability

The TLX NASA Usability Test revealed below-intermediate levels of perceived demand across categories, indicating that users did not find any task particularly challenging or demanding. Overall, the web application received an average positive rating of 5.5 out of 7, highlighting its potential as a valuable tool in the healthcare domain.1

Most (60%) survey respondents found the interface easy to navigate. Participants also reported an understanding of the complex data visualizations with minimal clarification needed.1 Popup tabs presenting additional data were rated highly self-explanatory by all users.

Users appreciated the self-explanatory information, detailed summary, and transparency of AI predictions. The detailed summary tab consolidating information scored best for ease of use, and participants appreciated the potential of this feature to streamline workflows. However, most users felt that patient demographics beyond age add little diagnostic value.

Furthermore, acceptability was high among the survey participants. All five participants expressed willingness to incorporate the application within their prostate cancer diagnostics, as they appreciated the tool’s potential to reduce their mental workload. Most participants did not express concerns related to AI bias.1

Suggestions focused on alternative, potentially more user-friendly chart layouts, including the addition of vertical bar charts to illustrate key histopathological characteristics.

“Some users suggested incorporating additional data points and higher-resolution biopsy images for more meaningful insights,” Prof. Amal added.


Limitations and Future Directions

Although positive feedback underscores this application’s alignment with clinical prostate cancer diagnosis, the sample size of the survey participants was small.

“Moving forward, the need for a more extensive and diverse participant pool would have benefits, and we aim to include over 20 expert participants from various global locations as solving clinical variability in a generalized manner can be challenging,” said Prof. Amal.

Aspects of improvement include chart representation, the need for more nuanced information, and presentation of AI results in a clear and concise manner that non-technical users can easily understand. Future efforts are also needed to refine the NLP summaries and explore integration with existing healthcare systems.

Ultimately, developing tools that integrate human expertise with AI-driven grading can transform clinical workflows and point-of-care cancer diagnostics. This study demonstrated the feasibility of incorporating advanced analytics into real-world patient care.

“The intention to continue developing better versions underscores the commitment to enhance user experience and clinical efficacy, and the insights gained from this study lay the foundation for future iterations and improvements. The need for compliance with healthcare data privacy regulations, obtaining necessary approvals, and providing user training further emphasizes the need for future studies to garner improved insights,” Prof. Amal concluded.



The study received no funding.



  1. Singh A, Randive S, Breggia A, Ahmad B, Christman R, Amal S. Enhancing Prostate Cancer Diagnosis with a Novel Artificial Intelligence-Based Web Application: Synergizing Deep Learning Models, Multimodal Data, and Insights from Usability Study with Pathologists. Cancers (Basel). 2023;15(23):5659. Published 2023 Nov 30. doi:10.3390/cancers15235659

Share This Post

Leave a Reply