Authors:

Farren M Martinus

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Deep learning models have the potential to augment laparoscopic surgery by overlaying segmented images onto the surgical field or providing real-time performance evaluations. Most existing research in the field of laparoscopic image analysis uses retrospective human data. With regard to prospective model evaluation and clinical translation, it is currently unknown to what extent computational models for laparoscopic image analysis trained on human data are transferable to data from large animal surgeries and vice versa. In this project, we are investigating the cross-species validation of laparoscopic image analysis models. Specifically, two laparoscopic image analysis models will be tested for their cross-species validity: a Convolutional Long Short- Term Memory (ConvLSTM) model for surgical tool detection and segmentation, and a Transformer-based segmentation model for organs and anatomical structures. For the surgical tool tracking model, public human (Cholec80) and porcine (Endoscopic Vision Challenge 2017) datasets are utilized. The anatomical segmentation and classification model will be tested using human and porcine datasets from the Dresden Surgical Anatomy Dataset (DSAD) and the Endoscopic Vision Challenge 2018, respectively. Effectiveness will be measured qualitatively through visual analysis of segmented video feeds and quantitatively using accuracy, precision, F1 score, recall, intersection over union (IOU), and mean average precision (mAP). In conclusion, we envision this work will provide an understanding of the relevance of animal testing for human applications and improve veterinary practices with AI tools developed on more abundant human data. Keywords: Deep Learning; Laparoscopy; Cross-Species

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Farren M Martinus