Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

Surgery Surgery

TEsoNet: knowledge transfer in surgical phase recognition from laparoscopic sleeve gastrectomy to the laparoscopic part of Ivor-Lewis esophagectomy.

In Surgical endoscopy ; h5-index 65.0

BACKGROUND : Surgical phase recognition using computer vision presents an essential requirement for artificial intelligence-assisted analysis of surgical workflow. Its performance is heavily dependent on large amounts of annotated video data, which remain a limited resource, especially concerning highly specialized procedures. Knowledge transfer from common to more complex procedures can promote data efficiency. Phase recognition models trained on large, readily available datasets may be extrapolated and transferred to smaller datasets of different procedures to improve generalizability. The conditions under which transfer learning is appropriate and feasible remain to be established.

METHODS : We defined ten operative phases for the laparoscopic part of Ivor-Lewis Esophagectomy through expert consensus. A dataset of 40 videos was annotated accordingly. The knowledge transfer capability of an established model architecture for phase recognition (CNN + LSTM) was adapted to generate a "Transferal Esophagectomy Network" (TEsoNet) for co-training and transfer learning from laparoscopic Sleeve Gastrectomy to the laparoscopic part of Ivor-Lewis Esophagectomy, exploring different training set compositions and training weights.

RESULTS : The explored model architecture is capable of accurate phase detection in complex procedures, such as Esophagectomy, even with low quantities of training data. Knowledge transfer between two upper gastrointestinal procedures is feasible and achieves reasonable accuracy with respect to operative phases with high procedural overlap.

CONCLUSION : Robust phase recognition models can achieve reasonable yet phase-specific accuracy through transfer learning and co-training between two related procedures, even when exposed to small amounts of training data of the target procedure. Further exploration is required to determine appropriate data amounts, key characteristics of the training procedure and temporal annotation methods required for successful transferal phase recognition. Transfer learning across different procedures addressing small datasets may increase data efficiency. Finally, to enable the surgical application of AI for intraoperative risk mitigation, coverage of rare, specialized procedures needs to be explored.

Eckhoff J A, Ban Y, Rosman G, Müller D T, Hashimoto D A, Witkowski E, Babic B, Rus D, Bruns C, Fuchs H F, Meireles O

2023-Mar-17

Artificial intelligence, Computer vision, Ivor–Lewis esophagectomy, Phase recognition, Transfer learning, Upper gastrointestinal surgery

Surgery Surgery

One-year review of real-time artificial intelligence (AI)-aided endoscopy performance.

In Surgical endoscopy ; h5-index 65.0

BACKGROUND : Colonoscopies have long been the gold standard for detection of pre-malignant neoplastic lesions of the colon. Our previous study tried real-time artificial intelligence (AI)-aided colonoscopy over a three-month period and found significant improvements in collective and individual endoscopist's adenoma detection rates compared to baseline. As an expansion, this study evaluates the 1-year performance of AI-aided colonoscopy in the same institution.

METHODS : A prospective cohort study was conducted in a single institution in Singapore. The AI software used was GI Genius™ Intelligent Endoscopy Module, US-DG-2000309 © 2021 Medtronic. Between July 2021 and June 2022, polypectomy rates in non-AI-aided colonoscopies and AI-aided colonoscopies were calculated and compared. Some of the AI-aided colonoscopies were recorded and video reviewed. A "hit" was defined as a sustained detection of an area by the AI. If a polypectomy was performed for a "hit," its histology was reviewed. Additional calculations for polyp detection rate (PDR), adenoma detection rate (ADR), and adenoma detection per colonoscopy (ADPC) were performed. Cost analysis was performed to determine cost effectiveness of subscription to the AI program.

RESULTS : 2433 AI-aided colonoscopies were performed between July 2021 and June 2022 and compared against 1770 non-AI-aided colonoscopies. AI-aided colonoscopies yielded significantly higher rates of polypectomies (33.6%) as compared with non-AI-aided colonoscopies (28.4%) (p < 0.001). Among the AI-aided colonoscopies, 1050 were reviewed and a final 843 were included for additional analysis. The polypectomy to "hit" ratio was 57.4%, PDR = 45.6%, ADR = 32.4%, and ADPC = 2.08. Histological review showed that 25 polyps (3.13%) were sessile-serrated adenomas. Cost analysis found that the increased polypectomy rates in AI-aided colonoscopes led to an increase in revenue, which covered the subscription cost with an excess of USD 20,000.

CONCLUSION : AI-aided colonoscopy is a cost effective means of improving colonoscopy quality and may help advance colorectal cancer screening in Singapore.

Chin Shuen-Ern, Wan Fang-Ting, Ladlad Jasmine, Chue Koy-Min, Teo Eng-Kiong, Lin Cui-Li, Foo Fung-Joon, Koh Frederick H

2023-Mar-17

Adenoma detection, Artificial intelligence, Colonoscopy, Endoscopy, Polyp detection

Cardiology Cardiology

Generalizable and robust deep learning algorithm for atrial fibrillation diagnosis across geography, ages and sexes.

In NPJ digital medicine

To drive health innovation that meets the needs of all and democratize healthcare, there is a need to assess the generalization performance of deep learning (DL) algorithms across various distribution shifts to ensure that these algorithms are robust. This retrospective study is, to the best of our knowledge, an original attempt to develop and assess the generalization performance of a DL model for AF events detection from long term beat-to-beat intervals across geography, ages and sexes. The new recurrent DL model, denoted ArNet2, is developed on a large retrospective dataset of 2,147 patients totaling 51,386 h obtained from continuous electrocardiogram (ECG). The model's generalization is evaluated on manually annotated test sets from four centers (USA, Israel, Japan and China) totaling 402 patients. The model is further validated on a retrospective dataset of 1,825 consecutives Holter recordings from Israel. The model outperforms benchmark state-of-the-art models and generalized well across geography, ages and sexes. For the task of event detection ArNet2 performance was higher for female than male, higher for young adults (less than 61 years old) than other age groups and across geography. Finally, ArNet2 shows better performance for the test sets from the USA and China. The main finding explaining these variations is an impairment in performance in groups with a higher prevalence of atrial flutter (AFL). Our findings on the relative performance of ArNet2 across groups may have clinical implications on the choice of the preferred AF examination method to use relative to the group of interest.

Biton Shany, Aldhafeeri Mohsin, Marcusohn Erez, Tsutsui Kenta, Szwagier Tom, Elias Adi, Oster Julien, Sellal Jean Marc, Suleiman Mahmoud, Behar Joachim A

2023-Mar-17

Radiology Radiology

Automatic detection of punctate white matter lesions in infants using deep learning of composite images from two cases.

In Scientific reports ; h5-index 158.0

Punctate white matter lesions (PWMLs) in infants may be related to neurodevelopmental outcomes based on the location or number of lesions. This study aimed to assess the automatic detectability of PWMLs in infants on deep learning using composite images created from several cases. To create the initial composite images, magnetic resonance (MR) images of two infants with the most PWMLs were used; their PWMLs were extracted and pasted onto MR images of infants without abnormality, creating many composite PWML images. Deep learning models based on a convolutional neural network, You Only Look Once v3 (YOLOv3), were constructed using the training set of 600, 1200, 2400, and 3600 composite images. As a result, a threshold of detection probability of 20% and 30% for all deep learning model sets yielded a relatively high sensitivity for automatic PWML detection (0.908-0.957). Although relatively high false-positive detections occurred with the lower threshold of detection probability, primarily, in the partial volume of the cerebral cortex (≥ 85.8%), those can be easily distinguished from the white matter lesions. Relatively highly sensitive automatic detection of PWMLs was achieved by creating composite images from two cases using deep learning.

Sun Xuyang, Niwa Tetsu, Okazaki Takashi, Kameda Sadanori, Shibukawa Shuhei, Horie Tomohiko, Kazama Toshiki, Uchiyama Atsushi, Hashimoto Jun

2023-Mar-17

General General

Opposite effects of positive and negative symptoms on resting-state brain networks in schizophrenia.

In Communications biology

Schizophrenia is a severe psychotic disorder characterized by positive and negative symptoms, but their neural bases remain poorly understood. Here, we utilized a nested-spectral partition (NSP) approach to detect hierarchical modules in resting-state brain functional networks in schizophrenia patients and healthy controls, and we studied dynamic transitions of segregation and integration as well as their relationships with clinical symptoms. Schizophrenia brains showed a more stable integrating process and a more variable segregating process, thus maintaining higher segregation, especially in the limbic system. Hallucinations were associated with higher integration in attention systems, and avolition was related to a more variable segregating process in default-mode network (DMN) and control systems. In a machine-learning model, NSP-based features outperformed graph measures at predicting positive and negative symptoms. Multivariate analysis confirmed that positive and negative symptoms had opposite effects on dynamic segregation and integration of brain networks. Gene ontology analysis revealed that the effect of negative symptoms was related to autistic, aggressive and violent behavior; the effect of positive symptoms was associated with hyperammonemia and acidosis; and the interaction effect was correlated with abnormal motor function. Our findings could contribute to the development of more accurate diagnostic criteria for positive and negative symptoms in schizophrenia.

Wang Xinrui, Chang Zhao, Wang Rong

2023-Mar-17

General General

Information set supported deep learning architectures for improving noisy image classification.

In Scientific reports ; h5-index 158.0

Deep learning models have been widely used in many supervised learning applications. However, these models suffer from overfitting due to various types of uncertainty with deteriorating performance when facing data biases, class imbalance, or noise propagation. The Information-Set Deep learning (ISDL) architectures with four variants are developed by integrating information set theory and deep learning principles to address the critical problem of the absence of robust deep learning models. There is a description of the ISDL architectures, learning algorithms, and analytic workflows. The performance of the ISDL models and standard architectures is evaluated using a noise-corrupted benchmark dataset. The experimental results show that the ISDL models can efficiently handle noise-dominated uncertainty and outperform peer architectures.

Bhardwaj Saurabh, Wang Yizhi, Yu Guoqiang, Wang Yue

2023-Mar-17