Détection précise du cancer du sein avec l'IA : étude de l'algorithme Gallon Breast

Table of Contents:

1. Introduction
2. Importance of AI in Cancer Diagnosis
3. The Gallon Breast Algorithm
4. Objective of the Study
5. Study Design and Methodology 5.1 Multi-site Study Locations 5.2 Study Participants 5.3 Sample Size Calculation 5.4 Study Flow Chart
6. Results and Findings 6.1 Invasive Carcinoma Detection 6.2 Insight Carcinoma Detection 6.3 Distinguishing Lobular vs Ductal Invasive Carcinoma 6.4 Distinguishing Low-grade vs High-grade DCIS
7. Comparison with Previous Studies
8. Conclusion
9. Implementation of the AI Solution
10. References

Introduction

Breast cancer is the most common malignant disease, and accurate diagnosis plays a crucial role in determining the appropriate treatment plan. Computer-assisted diagnostic solutions, powered by artificial intelligence (AI), have shown promise in improving the accuracy and efficiency of breast biopsy diagnosis. In this article, we will discuss the results of a multi-feature AI-based solution for cancer diagnosis in breast biopsies. The study, entitled "A Multi-Feature AI-Based Solution for Cancer Diagnosis in Breast Biopsies: A Prospective Blinded Multi-Site Clinical Study," aims to clinically validate the performance of the Gallon Breast algorithm.

Importance of AI in Cancer Diagnosis

Computer-assisted diagnostic solutions can greatly support pathologists in accurately diagnosing breast biopsies. These solutions utilize AI algorithms to detect multiple cancer types, such as invasive and in situ carcinomas, as well as additional clinical features like angiolymphatic invasions or chills. However, before these solutions can be implemented in routine practice, they need to undergo rigorously blinded independent multi-site clinical studies to ensure their reliability and effectiveness.

The Gallon Breast Algorithm

The Gallon Breast algorithm, developed by Ibex Medical Analytics, is an AI-based diagnostic algorithm specifically designed to detect various types of cancer in breast biopsies. It aims to achieve high accuracy and performance in detecting invasive carcinoma, in situ carcinoma, and grading of DCIS. The algorithm's performance was validated in this prospective blinded multi-site clinical study.

Objective of the Study

The primary objective of the study was to clinically validate the performance of the Gallon Breast algorithm in detecting invasive carcinoma. The secondary and exploratory endpoints included the algorithm's performance in detecting DCIS and ADH, distinguishing between ductal and lobular invasive carcinomas, and distinguishing between low-grade and high-grade DCIS.

Study Design and Methodology

The study involved multiple sites, including institutions in France and Israel. It utilized a prospective study design with retrospectively collected cases. Six senior pathologists participated in the study, and the ground truth was established through majority agreement by at least two independent pathologists who reviewed the cases without knowledge of the initial diagnosis. The sample size was determined by a statistician, and the study flow involved creating a dataset, scanning and processing the images using the Gallon Breast algorithm, and performing analysis and results evaluation.

Multi-site Study Locations

The study was conducted at institutions in France and Macabee Health Care Services in Israel. This multi-site approach ensured the generalizability and robustness of the findings.

Study Participants

The study included a diverse group of six senior pathologists who reviewed the cases independently and contributed to establishing the ground truth for comparison with the algorithm's results.

Sample Size Calculation

A statistician calculated the sample size based on achieving 80% power at a two-sided 5% significance level. The sample size calculation aimed to ensure sufficient statistical power to validate the algorithm's sensitivity and specificity for invasive cancer detection.

Study Flow Chart

The study flow involved creating a dataset consisting of 436 cases, which encompassed invasive carcinoma, DCIS and ADH, and benign lesions. The dataset was processed using the Gallon Breast algorithm, which was blinded to the ground truth. The algorithm's results were then compared to the ground truth, and any discrepancies were reviewed by a third pathologist to obtain the final ground truth.

Results and Findings

The study demonstrated that the Gallon Breast algorithm performed exceptionally well in detecting invasive carcinoma. The algorithm achieved a sensitivity of 95.51% and specificity of 93.57%, with an area under the curve (AUC) of 0.990. It also exhibited high performance in detecting in situ carcinoma, with an AUC of 0.949. The algorithm accurately distinguished between lobular and ductal invasive carcinoma, with an AUC of 0.973. Furthermore, it showed a high accuracy in differentiating low-grade and high-grade DCIS, with an AUC of 0.921.

Comparison with Previous Studies

When compared to previously published AI studies, the Present study stands out due to its large sample size and inclusion of a high number of whole slide images. The Gallon Breast algorithm demonstrated higher accuracy, sensitivity, and specificity in detecting invasive carcinoma and in situ carcinoma compared to previous studies. The algorithm's performance in detecting other features, such as angiolymphatic invasions or chills, was also noteworthy.

Conclusion

This prospective blinded multi-site clinical study successfully validated the performance of the Gallon Breast AI-based solution for cancer diagnosis in breast biopsies. The algorithm demonstrated high accuracy and performance in detecting invasive and in situ carcinomas, as well as distinguishing between different histological types and grades of DCIS. The implementation of this AI solution in routine practice, as a Second read application, could significantly improve the quality control of breast biopsy diagnoses.

Implementation of the AI Solution

Considering the high accuracy and performance demonstrated by the Gallon Breast algorithm, its implementation as a second read application in routine practice could greatly enhance the diagnostic process of breast biopsies. This AI-powered solution has the potential to act as a reliable quality control tool, ensuring the accuracy and consistency of breast cancer diagnoses.

References

[Insert Relevant references here]

Highlights:

• The Gallon Breast algorithm, an AI-based solution for cancer diagnosis in breast biopsies, demonstrated exceptionally high accuracy and performance.
• The algorithm accurately detected invasive and in situ carcinomas, as well as distinguished between different histological types and grades of DCIS.
• This prospective multi-site clinical study validated the reliability and effectiveness of the Gallon Breast algorithm.
• The implementation of this AI solution in routine practice could improve the quality control of breast biopsy diagnoses.

FAQ:

Q: What is the Gallon Breast algorithm? A: The Gallon Breast algorithm is an AI-based diagnostic algorithm developed by Ibex Medical Analytics. It is designed to detect various types of cancer in breast biopsies with high accuracy and performance.

Q: What were the primary and secondary objectives of the study? A: The primary objective of the study was to clinically validate the performance of the Gallon Breast algorithm in detecting invasive carcinoma. The secondary objectives included evaluating the algorithm's performance in detecting DCIS and ADH, distinguishing between ductal and lobular invasive carcinomas, and distinguishing between low-grade and high-grade DCIS.

Q: How was the ground truth established in the study? A: The ground truth was established through majority agreement by at least two independent pathologists who reviewed the cases without knowledge of the initial diagnosis. Any discrepancies were reviewed by a third pathologist to obtain the final ground truth.

Q: How does the performance of the Gallon Breast algorithm compare to previous studies? A: The Gallon Breast algorithm demonstrated higher accuracy, sensitivity, and specificity in detecting invasive and in situ carcinomas compared to previous AI studies. The algorithm also performed well in detecting other features, such as angiolymphatic invasions or chills.

Q: How can the implementation of the AI solution improve routine practice? A: The implementation of the Gallon Breast algorithm as a second read application in routine practice can enhance the quality control of breast biopsy diagnoses, ensuring greater accuracy and consistency.