Artificial Intelligence-Driven Structurization of Diagnostic Information in Free-Text Pathology Reports

Synoptic Reporting by Summarizing Cancer Pathology Reports using Large Language Models

“…While a high performance was reported, the authors report this was limited to data elements specific to breast cancer pathology reports that have been standardized at their institute [17]. It can be seen that the existing state-of-the-art NLP based methods for extracting structured data from unstructured data suffer from being limited in the types of reports they can analyse [17,18], have the tendency to lose important semantic information [14], often require semi-structured data [15], or require large amounts of expensive expert labeled data [18].…”

“…Savova et al [13] developed an open-source tool DeepPhe to extract cancer phenotype data from electronic health records (EHR). The techniques from open information extraction and artificial intelligence (AI) tools are combined in a heuristic manner for structurizing pathology reports in [14]. This proposed heuristic method has a primary disadvantage in that it is ad-hoc and requires the extensive redevelopment of the heuristic for pathology reports from a new source [14].…”

Empowering digital pathology applications through explainable knowledge extraction tools

“…However, the manual analysis of data becomes an extremely time-consuming process since reports vary widely between institutions, might be written in languages other than English, contain noise, and do not present a standard structure. In this context, Natural Language Processing (NLP) methods are central 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 as they empower the efficient automatic processing of thousands of clinical reports and the extraction of key information for several downstream tasks, such as clinical note mining 9 , 10 and structuring, 11 risk prediction, 12 clinical decision-support, 13 and precision medicine retrieval. 14…”

An accessible, efficient, and accurate natural language processing method for extracting diagnostic data from pathology reports

“… 7 , 8 , 11 , 12 , 18 , 19 More complex machine learning algorithms require tedious manual annotation by expert reviewers as part of a training dataset, and require tuning of statistical models to identify different parts of the report. 6 , 16 , 20 Similarly to rule-based methods, the output from these machine learning algorithms may be in the form of annotated mark-up files which may not be immediately useful for pathology research. 6 , 17 , 21 , 22 , 23 In addition, availability of resources such as computing power and specialty-trained personnel may limit their feasibility and usefulness.…”