Back

Pulmonary Hypertension Classification using Artificial Intelligence and Chest X-Ray:ATA AI STUDY-1

Kivrak, T.; Yagmur, B.; Erken, H.; Tuncer, T.; Dogan, S.; Yaman, O.; Sinan, U. Y.; Sert Sekerci, S.; Yayla, C.; Iyigun, U.; Kis, M.; Karaca, O.; Yesil, E.; Yuce Ersoy, E. I.; Tekin Tak, B.; Oz, A.; Kaplan, M.; Ulutas, Z.; Aslan, G. Y.; Kahya Eren, N.; Turhan Caglar, F. N.; Solmaz, H.; Ozden, O.; Gunes, H.; Kocabas, U.; Yenercag, M.; Isik, O.; Yesilkaya, C.; Kaya, A. N.; Omur, S. E.; Sahin, A.; In, E.; Kırıcı Berber, N.; Ileri Dogan, C.; Poyraz, F.; Kaya, E. E.; Gumusdag, A.; Kumet, O.; Kaya, H.; Sarikaya, R.; Tan, S. T.; Arabaci, H. O.; Cetin Guvenc, R.; Yeni, M.; Kilickiran Avci, B.; Yilmaz,

2023-04-17 cardiovascular medicine
10.1101/2023.04.14.23288561 medRxiv
Show abstract

An accurate diagnosis of pulmonary hypertension (PH) is crucial to ensure that patients receive timely treatment. One of the used imaging models to detect pulmonary hypertension is the X-ray. Therefore, a new automated PH-type classification model has been presented to depict the separation ability of deep learning for PH types. We retrospectively enrolled 6642 images of patients with PH and the control group. A new X-ray image dataset was collected from a multicentre in this work. A transfer learning-based image classification model has been presented in classifying PH types. Our proposed model was applied to the collected dataset, and this dataset contains six categories (five PH and a non-PH). The presented deep feature engineering (computer vision) model attained 86.14% accuracy on this dataset. According to the extracted ROC curve, the average area under the curve rate has been calculated at 0.945. Therefore, we believe that our proposed model can easily separate PH and non-PH X-ray images.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.